{ "version": "2026-09-29", "description": "Normalised edge AI inference benchmark database: one row per hardware × model × runtime × precision × input × power mode, deduplicated with the best-evidenced row kept. Class C rows are measured and published by the source named on the row (Ultralytics, Google Coral, Hailo Model Zoo, NVIDIA-AI-IOT, MLCommons, Jetson AI Lab, hardware vendors such as Seeed Studio, peer-reviewed papers, and community benchmarks that publish their method and raw results); class D rows are EdgeAIStack-derived estimates with the derivation stated. Metrics: fps, latency (p50/p99), tokens/s, power, memory, accuracy where published — latency_source says whether a latency was observed or computed as 1000/fps, and every p99 is a modelled tail rather than an observed one.", "last_updated": "2026-09-16", "confidence": "medium", "evidence_class": "C/D", "stats": { "rows": 605, "by_class": { "C": 310, "D": 295 }, "by_origin": { "research_corpus": 113, "catalog": 480, "p185": 12 }, "by_runtime": { "edge_tpu": 40, "hailo_sdk": 134, "tensorrt": 357, "mlc": 4, "pytorch": 5, "unspecified": 8, "openvino": 25, "cpu": 5, "rknn": 27 }, "by_precision": { "int8": 327, "fp16": 213, "int4": 4, "fp32": 32, "unspecified": 8, "fp8": 21 }, "by_task": { "segmentation": 75, "classification": 61, "detection": 395, "pose": 52, "text_generation": 8, "lidar_detection_3d": 2, "speech_recognition": 8, "vision_language": 4 }, "hardware": 24, "families": 49, "deduped": 96, "by_latency_source": { "derived_reciprocal": 572, "measured": 21 }, "by_latency_p99_source": { "derived": 113 } }, "documents": { "ultralytics_jetson": { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "ultralytics_rpi": { "publisher": "Ultralytics", "document": "Raspberry Pi guide — YOLO benchmarks", "doc_id": "ultralytics:guides/raspberry-pi", "url": "https://docs.ultralytics.com/guides/raspberry-pi/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "ultralytics_yolov8": { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "ultralytics_yolo11": { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "ultralytics_yolov10": { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "ultralytics_rtdetr": { "publisher": "Ultralytics", "document": "RT-DETR model page — performance table", "doc_id": "ultralytics:models/rtdetr", "url": "https://docs.ultralytics.com/models/rtdetr/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "coral_benchmarks": { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "coral_zoo_cls": { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "coral_zoo_det": { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "automl_effdet": { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo8_det": { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 object detection", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo8_cls": { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo8l_det": { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8L object detection", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8L_object_detection", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo8l_cls": { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8L classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8L_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo_rpi5": { "publisher": "Hailo (community examples)", "document": "hailo-rpi5-examples benchmark README", "doc_id": "github:hailo-ai/hailo-rpi5-examples", "url": "https://github.com/hailo-ai/hailo-rpi5-examples", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "nanosam": { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "nanosam README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/nanosam", "url": "https://github.com/NVIDIA-AI-IOT/nanosam", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "whisper_trt": { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "whisper_trt README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/whisper_trt", "url": "https://github.com/NVIDIA-AI-IOT/whisper_trt", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "mlperf_edge": { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "hailo_sdk": { "publisher": "Hailo", "document": "Hailo-8 SDK benchmarks (Model Zoo)", "doc_id": "github:hailo-ai/hailo_model_zoo", "url": "https://github.com/hailo-ai/hailo_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16" }, "corpus": { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16" } }, "hardware": { "coral_tpu": { "id": "coral_tpu", "label": "Google Coral Edge TPU", "in_catalog": true, "rows": 40 }, "hailo_10h": { "id": "hailo_10h", "label": "Hailo-10H", "in_catalog": true, "rows": 36 }, "hailo_8": { "id": "hailo_8", "label": "Hailo-8", "in_catalog": true, "rows": 52 }, "hailo_8l": { "id": "hailo_8l", "label": "Hailo-8L", "in_catalog": true, "rows": 46 }, "jetson_agx_orin": { "id": "jetson_agx_orin", "label": "Jetson AGX Orin 64GB", "in_catalog": true, "rows": 64 }, "jetson_agx_orin_32gb": { "id": "jetson_agx_orin_32gb", "label": "Jetson AGX Orin 32GB", "in_catalog": true, "rows": 16 }, "jetson_orin_nano": { "id": "jetson_orin_nano", "label": "Jetson Orin Nano 8GB", "in_catalog": true, "rows": 18 }, "jetson_orin_nano_4gb": { "id": "jetson_orin_nano_4gb", "label": "Jetson Orin Nano 4GB", "in_catalog": false, "rows": 6 }, "jetson_orin_nano_super": { "id": "jetson_orin_nano_super", "label": "Jetson Orin Nano Super", "in_catalog": true, "rows": 63 }, "jetson_orin_nx": { "id": "jetson_orin_nx", "label": "Jetson Orin NX 16GB", "in_catalog": true, "rows": 50 }, "jetson_orin_nx_8gb": { "id": "jetson_orin_nx_8gb", "label": "Jetson Orin NX 8GB", "in_catalog": true, "rows": 11 }, "jetson_thor_t5000": { "id": "jetson_thor_t5000", "label": "Jetson Thor T5000", "in_catalog": true, "rows": 90 }, "neousys_nuvo9531": { "id": "neousys_nuvo9531", "label": "Neousys Nuvo-9531 (Intel 12th-14th Gen Core)", "in_catalog": true, "rows": 25 }, "nvidia_a100": { "id": "nvidia_a100", "label": "NVIDIA A100 (server)", "in_catalog": false, "rows": 9 }, "nvidia_a10_server": { "id": "nvidia_a10_server", "label": "NVIDIA A10 (24GB)", "in_catalog": true, "rows": 7 }, "nvidia_a2_server": { "id": "nvidia_a2_server", "label": "NVIDIA A2 (16GB)", "in_catalog": true, "rows": 5 }, "nvidia_l40s_server": { "id": "nvidia_l40s_server", "label": "NVIDIA L40S (48GB)", "in_catalog": true, "rows": 4 }, "nvidia_l4_server": { "id": "nvidia_l4_server", "label": "NVIDIA L4 (24GB)", "in_catalog": true, "rows": 9 }, "nvidia_t4": { "id": "nvidia_t4", "label": "NVIDIA T4 (server)", "in_catalog": false, "rows": 17 }, "nvidia_v100": { "id": "nvidia_v100", "label": "NVIDIA V100 (server)", "in_catalog": false, "rows": 5 }, "raspberry_pi_5": { "id": "raspberry_pi_5", "label": "Raspberry Pi 5 (CPU)", "in_catalog": false, "rows": 5 }, "rk3568": { "id": "rk3568", "label": "Rockchip RK3568", "in_catalog": true, "rows": 1 }, "rk3588_npu": { "id": "rk3588_npu", "label": "Rockchip RK3588 NPU", "in_catalog": true, "rows": 25 }, "rk3588s": { "id": "rk3588s", "label": "Rockchip RK3588S", "in_catalog": true, "rows": 1 } }, "schema": { "rows[]": "{ id, key, origin (catalog | research_corpus | p185), hardware, power_mode, software {name, version}, model {family, series, variant, task}, input {resolution, batch}, runtime, runtime_inferred, precision, accelerator (gpu | igpu | npu | tpu | cpu), scope (inference_only | end_to_end), metrics {fps, latency_ms, latency_p99_ms, tokens_per_s, power_w, memory_mb, accuracy_map50_95}, class (C | D), confidence, date, derivation, notes, sources[], duplicates[], latency_source, latency_p99_source, latency_p99_basis }", "class": "C = measured and published by the source named in sources[]; D = EdgeAIStack model (derivation states the scaling used)", "duplicates[]": "other rows that described the same configuration before deduplication, with their origin, class, value and source id", "latency_source": "measured = metrics.latency_ms was observed; derived_reciprocal = it equals 1000/fps to within 2 % and was computed from the throughput, not measured; null = the row carries no latency. Almost every row that reports both metrics is derived_reciprocal — read this flag before presenting a latency as a measurement.", "latency_p99_source": "derived = metrics.latency_p99_ms is a modelled 1.20–1.43× multiplier over the p50 (every p99 in this database is); measured = an observed tail; null = no p99 on the row.", "latency_p99_basis": "When latency_p99_source is derived, names the multiplier band used at build time (see build-benchmarks-v2.js)." }, "rows": [ { "origin": "research_corpus", "record_id": "coral-deeplab-v3-tpu-int8-159", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "deeplab", "series": "deeplab", "variant": "v3", "task": "segmentation" }, "input": { "resolution": "513x513", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 4.1, "latency_ms": 241, "latency_p99_ms": 300, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. DeepLabV3 semantic segmentation. Computationally heavy; partial off-chip execution.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|deeplab|v3|edge_tpu|int8|513x513|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0575-coral-tpu-deeplab-v3-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "densenet", "series": "densenet", "variant": "121", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 50, "latency_ms": 20, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "key": "coral_tpu|densenet|121|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0330-coral-tpu-densenet-121-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientdet-lite0-coral-usb-int8-063", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "lite0", "task": "detection" }, "input": { "resolution": "320x320", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 26.7, "latency_ms": 37.4, "latency_p99_ms": 45, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.304 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator on desktop host CPU, EdgeTPU INT8 quantized", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet|lite0|edge_tpu|int8|320x320|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0506-coral-tpu-efficientdet-lite0-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientdet-lite1-coral-usb-int8-064", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "lite1", "task": "detection" }, "input": { "resolution": "384x384", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 17.8, "latency_ms": 56.3, "latency_p99_ms": 68, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.343 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator on desktop host CPU, EdgeTPU INT8 quantized", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet|lite1|edge_tpu|int8|384x384|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0507-coral-tpu-efficientdet-lite1-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientdet-lite2-coral-usb-int8-065", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "lite2", "task": "detection" }, "input": { "resolution": "448x448", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 9.6, "latency_ms": 104.6, "latency_p99_ms": 126, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.36 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator on desktop host CPU, EdgeTPU INT8 quantized", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet|lite2|edge_tpu|int8|448x448|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0508-coral-tpu-efficientdet-lite2-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientdet-lite3-coral-usb-int8-066", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "lite3", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 9.3, "latency_ms": 107.6, "latency_p99_ms": 130, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.394 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator on desktop host CPU, EdgeTPU INT8 quantized", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet|lite3|edge_tpu|int8|512x512|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0509-coral-tpu-efficientdet-lite3-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "efficientdet_lite", "series": "efficientdet_lite", "variant": "lite0", "task": "detection" }, "input": { "resolution": "320x320", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 98, "latency_ms": 10.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet_lite|lite0|edge_tpu|int8|320x320|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0053-coral-tpu-efficientdet-lite-lite0-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "efficientdet_lite", "series": "efficientdet_lite", "variant": "lite1", "task": "detection" }, "input": { "resolution": "384x384", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 65, "latency_ms": 15.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet_lite|lite1|edge_tpu|int8|384x384|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0054-coral-tpu-efficientdet-lite-lite1-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "efficientdet_lite", "series": "efficientdet_lite", "variant": "lite2", "task": "detection" }, "input": { "resolution": "448x448", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 50, "latency_ms": 20, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet_lite|lite2|edge_tpu|int8|448x448|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0055-coral-tpu-efficientdet-lite-lite2-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "efficientdet_lite", "series": "efficientdet_lite", "variant": "lite3", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 26, "latency_ms": 38.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientdet_lite|lite3|edge_tpu|int8|512x512|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0056-coral-tpu-efficientdet-lite-lite3-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientdet_lite", "series": "efficientdet_lite", "variant": "lite3x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 14, "latency_ms": 70, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "Derived, not measured: Coral publishes no EfficientDet-Lite3x benchmark. EfficientDet-Lite3x (640x640) exists as EdgeTPU-compiled model in google-coral/test_data but has NO official benchmark. Latency estimated from Lite3 (512x512, 38.5ms) scaled by resolution ratio (640/512)^2 * architectural overhead = ~1.82x -> 70ms. Confidence downgraded to medium.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "key": "coral_tpu|efficientdet_lite|lite3x|edge_tpu|int8|640x640|1|usb3", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0665-coral-tpu-efficientdet-lite-lite3x-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientnet-edgetpu-l-coral-usb-int8-076", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet", "series": "efficientnet", "variant": "edgetpu-l", "task": "classification" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 46.9, "latency_ms": 21.3, "latency_p99_ms": 26, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU-optimized EfficientNet, ImageNet Top-1=81.2%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet|edgetpu-l|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0518-coral-tpu-efficientnet-edgetpu-l-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientnet-edgetpu-m-coral-usb-int8-075", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet", "series": "efficientnet", "variant": "edgetpu-m", "task": "classification" }, "input": { "resolution": "240x240", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 137, "latency_ms": 7.3, "latency_p99_ms": 9, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU-optimized EfficientNet, ImageNet Top-1=80.1%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet|edgetpu-m|edge_tpu|int8|240x240|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0517-coral-tpu-efficientnet-edgetpu-m-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "efficientnet-edgetpu-s-coral-usb-int8-074", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet", "series": "efficientnet", "variant": "edgetpu-s", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 200, "latency_ms": 5, "latency_p99_ms": 6.2, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU-optimized EfficientNet, ImageNet Top-1=78.9%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet|edgetpu-s|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0516-coral-tpu-efficientnet-edgetpu-s-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet_edgetpu", "series": "efficientnet_edgetpu", "variant": "l", "task": "classification" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 39, "latency_ms": 25.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Benchmark table", "doc_id": "bench:publisher-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet_edgetpu|l|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0324-coral-tpu-efficientnet-edgetpu-l-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet_edgetpu", "series": "efficientnet_edgetpu", "variant": "m", "task": "classification" }, "input": { "resolution": "240x240", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 114, "latency_ms": 8.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Benchmark table", "doc_id": "bench:publisher-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet_edgetpu|m|edge_tpu|int8|240x240|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0323-coral-tpu-efficientnet-edgetpu-m-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "efficientnet_edgetpu", "series": "efficientnet_edgetpu", "variant": "s", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 196, "latency_ms": 5.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Benchmark table", "doc_id": "bench:publisher-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnet_edgetpu|s|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0322-coral-tpu-efficientnet-edgetpu-s-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-efficientnet-edgetpu-l-int8-157", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "efficientnetedgetpu", "series": "efficientnet-edgetpu", "variant": "l", "task": "classification" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 32.8, "latency_ms": 30.5, "latency_p99_ms": 38, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. EfficientNet-EdgeTPU-L.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnetedgetpu|l|edge_tpu|int8|300x300|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0573-coral-tpu-efficientnetedgetpu-l-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-efficientnet-edgetpu-m-int8-156", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "efficientnetedgetpu", "series": "efficientnet-edgetpu", "variant": "m", "task": "classification" }, "input": { "resolution": "240x240", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 94.3, "latency_ms": 10.6, "latency_p99_ms": 13.5, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. EfficientNet-EdgeTPU-M.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnetedgetpu|m|edge_tpu|int8|240x240|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0572-coral-tpu-efficientnetedgetpu-m-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-efficientnet-edgetpu-s-int8-155", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "efficientnetedgetpu", "series": "efficientnet-edgetpu", "variant": "s", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 181.8, "latency_ms": 5.5, "latency_p99_ms": 7, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. EfficientNet-EdgeTPU-S designed for full Edge TPU mapping.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|efficientnetedgetpu|s|edge_tpu|int8|224x224|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0571-coral-tpu-efficientnetedgetpu-s-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "inception-v1-coral-usb-int8-073", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "inception", "series": "inception", "variant": "v1", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 294.1, "latency_ms": 3.4, "latency_p99_ms": 4.2, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, ImageNet Top-1=71.9%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|inception|v1|edge_tpu|int8|224x224|1|usb3", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 294, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0327-coral-tpu-inception-v1-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-inception-v1-tpu-int8-160", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "inception", "series": "inception", "variant": "v1", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 243.9, "latency_ms": 4.1, "latency_p99_ms": 5.2, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. Inception v1 classification.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|inception|v1|edge_tpu|int8|224x224|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0576-coral-tpu-inception-v1-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "inception", "series": "inception", "variant": "v4", "task": "classification" }, "input": { "resolution": "299x299", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 11, "latency_ms": 85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "key": "coral_tpu|inception|v4|edge_tpu|int8|299x299|1|usb3", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0328-coral-tpu-inception-v4-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "ssd-mobilenetv1-coral-usb-int8-067", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "ssd-mobilenet-v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 153.8, "latency_ms": 6.5, "latency_p99_ms": 8, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.215 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, COCO mAP", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|ssd-mobilenet-v1|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0510-coral-tpu-mobilenet-ssd-mobilenet-v1-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "ssd-mobilenetv2-coral-usb-int8-068", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "ssd-mobilenet-v2", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 136.9, "latency_ms": 7.3, "latency_p99_ms": 9, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": 0.256 }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, COCO mAP", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — object detection", "doc_id": "coral:models/object-detection", "url": "https://coral.ai/models/object-detection/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|ssd-mobilenet-v2|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0511-coral-tpu-mobilenet-ssd-mobilenet-v2-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 417, "latency_ms": 2.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v1|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0046-coral-tpu-mobilenet-v1-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "mobilenetv1-coral-usb-int8-069", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1-1.0", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 357.1, "latency_ms": 2.8, "latency_p99_ms": 3.5, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, ImageNet Top-1=69.5%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v1-1.0|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0512-coral-tpu-mobilenet-v1-1-0-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1-ssd", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 154, "latency_ms": 6.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v1-ssd|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0048-coral-tpu-mobilenet-v1-ssd-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v2", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 385, "latency_ms": 2.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v2|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0047-coral-tpu-mobilenet-v2-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "mobilenetv2-coral-usb-int8-070", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v2-1.0", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 344.8, "latency_ms": 2.9, "latency_p99_ms": 3.6, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, ImageNet Top-1=73.2%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v2-1.0|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0513-coral-tpu-mobilenet-v2-1-0-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "mobilenetv3-coral-usb-int8-071", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v3", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 333.3, "latency_ms": 3, "latency_p99_ms": 3.8, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, ImageNet Top-1=77.5%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenet|v3|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0514-coral-tpu-mobilenet-v3-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-mobilenet-v1-ssd-tpu-int8-153", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "mobilenetssd", "series": "mobilenet-ssd", "variant": "mobilenet-v1-ssd", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 90.9, "latency_ms": 11, "latency_p99_ms": 14, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board M.2 (Edge TPU, 4 TOPS). INT8 TFLite. MobileNet v1 SSD object detection.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenetssd|mobilenet-v1-ssd|edge_tpu|int8|224x224|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0569-coral-tpu-mobilenetssd-mobilenet-v1-ssd-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-mobilenet-v2-ssd-tpu-int8-154", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "mobilenetssd", "series": "mobilenet-ssd", "variant": "mobilenet-v2-ssd", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 71.4, "latency_ms": 14, "latency_p99_ms": 18, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board M.2. INT8 TFLite. MobileNet v2 SSD.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|mobilenetssd|mobilenet-v2-ssd|edge_tpu|int8|224x224|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0570-coral-tpu-mobilenetssd-mobilenet-v2-ssd-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "resnet50-coral-usb-int8-072", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 23.7, "latency_ms": 42.2, "latency_p99_ms": 51, "tokens_per_s": null, "power_w": 1, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2022-Q2", "derivation": null, "notes": "Coral USB Accelerator, EdgeTPU INT8, ImageNet Top-1=73.6%", "sources": [ { "publisher": "Google Coral", "document": "Model zoo — image classification", "doc_id": "coral:models/image-classification", "url": "https://coral.ai/models/image-classification/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|resnet|50|edge_tpu|int8|224x224|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0515-coral-tpu-resnet-50-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-resnet50-v1-tpu-int8-161", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50-v1", "task": "classification" }, "input": { "resolution": "299x299", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 17.9, "latency_ms": 56, "latency_p99_ms": 70, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. ResNet-50 at 299x299. Slow due to residual connections not fully mapping to Edge TPU ops.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|resnet|50-v1|edge_tpu|int8|299x299|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0577-coral-tpu-resnet-50-v1-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50-v2", "task": "classification" }, "input": { "resolution": "299x299", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 20, "latency_ms": 50, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Benchmark table", "doc_id": "bench:publisher-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|resnet|50-v2|edge_tpu|int8|299x299|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0329-coral-tpu-resnet-50-v2-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v1-fpn", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 4, "latency_ms": 250, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|ssd_mobilenet|v1-fpn|edge_tpu|int8|640x640|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0058-coral-tpu-ssd-mobilenet-v1-fpn-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": { "name": "sdk", "version": "coral-2021" }, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v2", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 138, "latency_ms": 7.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2021", "derivation": null, "notes": null, "sources": [ { "publisher": "Google", "document": "Publisher benchmark table", "doc_id": "bench:google-publisher-benchmark-table", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|ssd_mobilenet|v2|edge_tpu|int8|300x300|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0057-coral-tpu-ssd-mobilenet-v2-edge-tpu-int8" }, { "origin": "catalog", "hardware": "coral_tpu", "power_mode": "usb3", "software": null, "model": { "family": "ssdlite_mobiledet", "series": "ssdlite_mobiledet", "variant": "edgetpu", "task": "detection" }, "input": { "resolution": "320x320", "batch": 1 }, "runtime": "edge_tpu", "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 65, "latency_ms": 15.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "key": "coral_tpu|ssdlite_mobiledet|edgetpu|edge_tpu|int8|320x320|1|usb3", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0326-coral-tpu-ssdlite-mobiledet-edgetpu-edge-tpu-int8" }, { "origin": "research_corpus", "record_id": "coral-unet-mv2-tpu-int8-158", "hardware": "coral_tpu", "power_mode": "m2", "software": null, "model": { "family": "unet", "series": "unet", "variant": "mv2", "task": "segmentation" }, "input": { "resolution": "128x128", "batch": 1 }, "runtime": "edge_tpu", "runtime_inferred": true, "precision": "int8", "accelerator": "tpu", "scope": "inference_only", "metrics": { "fps": 175.4, "latency_ms": 5.7, "latency_p99_ms": 7.2, "tokens_per_s": null, "power_w": 2, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q1", "derivation": null, "notes": "Coral Dev Board Edge TPU. UNet MobileNetV2 backbone, 128x128 input.", "sources": [ { "publisher": "Google Coral", "document": "Edge TPU benchmarks", "doc_id": "coral:docs/edgetpu/benchmarks", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "coral_tpu|unet|mv2|edge_tpu|int8|128x128|1|m2", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0574-coral-tpu-unet-mv2-edge-tpu-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "centernet", "series": "centernet", "variant": "resnet18", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 240, "latency_ms": 4.17, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|centernet|resnet18|hailo_sdk|int8|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0231-hailo-10h-centernet-resnet18-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 2864, "latency_ms": 0.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|mobilenet|v1|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0228-hailo-10h-mobilenet-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 2944, "latency_ms": 0.34, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|mobilenetv2|v2|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0227-hailo-10h-mobilenetv2-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 316, "latency_ms": 3.16, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|resnet|50|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0226-hailo-10h-resnet-50-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 501, "latency_ms": 2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|ssd_mobilenet|v1|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0224-hailo-10h-ssd-mobilenet-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v2", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 498, "latency_ms": 2.01, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|ssd_mobilenet|v2|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0225-hailo-10h-ssd-mobilenet-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 41.1, "latency_ms": 24.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0221-hailo-10h-yolo11-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 70.3, "latency_ms": 14.22, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0220-hailo-10h-yolo11-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 302, "latency_ms": 3.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 304, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0218-hailo-10h-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 332, "latency_ms": 3.01, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0468-hailo-10h-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 142, "latency_ms": 7.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 142, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0219-hailo-10h-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 223, "latency_ms": 4.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolo11|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0469-hailo-10h-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 303, "latency_ms": 3.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov10|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0471-hailo-10h-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 345, "latency_ms": 2.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov10|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0472-hailo-10h-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 146, "latency_ms": 6.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov10|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0473-hailo-10h-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 230, "latency_ms": 4.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov10|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0474-hailo-10h-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 112, "latency_ms": 8.93, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0230-hailo-10h-yolov5-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 250, "latency_ms": 4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 249, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0229-hailo-10h-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 296, "latency_ms": 3.38, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0470-hailo-10h-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 43.2, "latency_ms": 23.15, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5_seg|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0295-hailo-10h-yolov5-seg-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 65.7, "latency_ms": 15.22, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0294-hailo-10h-yolov5-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 249, "latency_ms": 4.02, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0292-hailo-10h-yolov5-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 177, "latency_ms": 5.65, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov5_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0293-hailo-10h-yolov5-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 41.6, "latency_ms": 24.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0287-hailo-10h-yolov8-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 76.2, "latency_ms": 13.12, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|m|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 75.1, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0286-hailo-10h-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 132, "latency_ms": 7.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|m|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0467-hailo-10h-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 375, "latency_ms": 2.67, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 378, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0222-hailo-10h-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 370, "latency_ms": 2.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0465-hailo-10h-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 166, "latency_ms": 6.02, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 168, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0223-hailo-10h-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v5.4.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 252, "latency_ms": 3.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO10H/HAILO10H_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0466-hailo-10h-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 25.6, "latency_ms": 39.06, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0288-hailo-10h-yolov8-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 79.2, "latency_ms": 12.63, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8_pose|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0297-hailo-10h-yolov8-pose-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 161, "latency_ms": 6.21, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8_pose|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0296-hailo-10h-yolov8-pose-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 65.6, "latency_ms": 15.24, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0291-hailo-10h-yolov8-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 310, "latency_ms": 3.23, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0289-hailo-10h-yolov8-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_10h", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 140, "latency_ms": 7.14, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_10h|yolov8_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0290-hailo-10h-yolov8-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "centernet", "series": "centernet", "variant": "resnet18", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 405, "latency_ms": 2.47, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|centernet|resnet18|hailo_sdk|int8|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0201-hailo-8-centernet-resnet18-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "centernet", "series": "centernet", "variant": "resnet50", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 87.1, "latency_ms": 11.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|centernet|resnet50|hailo_sdk|int8|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0202-hailo-8-centernet-resnet50-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "detr", "series": "detr", "variant": "resnet18", "task": "detection" }, "input": { "resolution": "800x800", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 30.5, "latency_ms": 32.79, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|detr|resnet18|hailo_sdk|int8|800x800|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0193-hailo-8-detr-resnet18-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8-efficientdet-d0-int8-140", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 95, "latency_ms": 10.5, "latency_p99_ms": 13.5, "tokens_per_s": null, "power_w": 3.5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8 26 TOPS, batch=8. EfficientDet-D0.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 object detection", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|efficientdet|d0|hailo_sdk|int8|512x512|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0563-hailo-8-efficientdet-d0-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 3305, "latency_ms": 0.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|mobilenet|v1|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0215-hailo-8-mobilenet-v1-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8-mobilenetv2-int8-138", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1100, "latency_ms": 0.9, "latency_p99_ms": 1.2, "tokens_per_s": null, "power_w": 2.5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8 26 TOPS, batch=8. Classification extremely efficient on Hailo.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|mobilenet|v2|hailo_sdk|int8|224x224|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0561-hailo-8-mobilenet-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "mobilenet_ssd", "series": "mobilenet_ssd", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 800, "latency_ms": 1.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|mobilenet_ssd|v1|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0137-hailo-8-mobilenet-ssd-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 2597, "latency_ms": 0.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · Hailo Model Zoo classification table, mobilenet_v2_1.0: FPS (Batch Size=1) 2597, FPS (Batch Size=8) 2597; top-1 71.8 float, 70.9 on hardware. System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x 4 lanes, room temperature. Hailo Dataflow Compiler v2.19.0.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo classification", "doc_id": "bench:hailo-model-zoo-classification", "revision": "v2.19.0", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_classification.rst", "retrieved": "2026-09-29", "verified": "2026-09-29", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|mobilenetv2|v2|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0612-hailo-8-mobilenetv2-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1372, "latency_ms": 0.73, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|resnet|50|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0195-hailo-8-resnet-50-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8-resnet50-int8-139", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 350, "latency_ms": 2.9, "latency_p99_ms": 3.7, "tokens_per_s": null, "power_w": 4.5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8 26 TOPS, batch=8. ResNet-50 classification baseline.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|resnet|50|hailo_sdk|int8|224x224|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0562-hailo-8-resnet-50-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1015, "latency_ms": 0.99, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|ssd_mobilenet|v1|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0189-hailo-8-ssd-mobilenet-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v2", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 784, "latency_ms": 1.28, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|ssd_mobilenet|v2|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0190-hailo-8-ssd-mobilenet-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 38, "latency_ms": 26.32, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|l|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 27, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0060-hailo-8-yolo11-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": "m2", "software": { "name": "sdk", "version": "sdk-4.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 55, "latency_ms": 18.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Hailo-8", "document": "Publisher benchmark table", "doc_id": "bench:hailo-8-publisher-benchmark-table", "url": "https://hailo.ai/developer-zone/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "key": "hailo_8|yolo11|m|hailo_sdk|int8|640x640|1|m2", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 44, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "catalog", "class": "D", "fps": 50.2, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0050-hailo-8-yolo11-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 185, "latency_ms": 5.41, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 108, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "catalog", "class": "C", "fps": 185, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0059-hailo-8-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 541, "latency_ms": 1.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0458-hailo-8-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 111, "latency_ms": 9.01, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|s|hailo_sdk|int8|640x640|1|m2", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 72, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "catalog", "class": "C", "fps": 105, "tokens_per_s": null, "source": "bench:hailo-8-sdk-benchmarks" }, { "origin": "catalog", "class": "C", "fps": 111, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0049-hailo-8-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 303, "latency_ms": 3.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0459-hailo-8-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 14.2, "latency_ms": 70.42, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0275-hailo-8-yolo11-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 47.7, "latency_ms": 20.96, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_pose|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0347-hailo-8-yolo11-pose-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 167, "latency_ms": 5.99, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_pose|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0345-hailo-8-yolo11-pose-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 105, "latency_ms": 9.52, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_pose|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0346-hailo-8-yolo11-pose-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 38.2, "latency_ms": 26.18, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0339-hailo-8-yolo11-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 130, "latency_ms": 7.69, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0337-hailo-8-yolo11-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 80, "latency_ms": 12.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolo11_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0338-hailo-8-yolo11-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 194, "latency_ms": 5.15, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov10|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0461-hailo-8-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 567, "latency_ms": 1.76, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov10|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0462-hailo-8-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 117, "latency_ms": 8.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov10|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0463-hailo-8-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 283, "latency_ms": 3.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov10|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0464-hailo-8-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 156, "latency_ms": 6.41, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0198-hailo-8-yolov5-m-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8-yolov5m-int8-137", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 85, "latency_ms": 11.8, "latency_p99_ms": 15, "tokens_per_s": null, "power_w": 5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8 26 TOPS, HailoRT, batch=8.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8 object detection", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5|m|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0560-hailo-8-yolov5-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 543, "latency_ms": 1.84, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 543, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0197-hailo-8-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 543, "latency_ms": 1.84, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5|s|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 140, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0460-hailo-8-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 69, "latency_ms": 14.49, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0211-hailo-8-yolov5-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 464, "latency_ms": 2.16, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0209-hailo-8-yolov5-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 331, "latency_ms": 3.02, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov5_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0210-hailo-8-yolov5-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 36.3, "latency_ms": 27.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0284-hailo-8-yolov8-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 66.9, "latency_ms": 14.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|m|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 67.3, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0283-hailo-8-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 149, "latency_ms": 6.71, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|m|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 45, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0457-hailo-8-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1036, "latency_ms": 0.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 1036, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0281-hailo-8-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1036, "latency_ms": 0.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|n|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 385, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "research_corpus", "class": "C", "fps": 107, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0455-hailo-8-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 491, "latency_ms": 2.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 491, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0282-hailo-8-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 491, "latency_ms": 2.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|s|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 200, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "research_corpus", "class": "C", "fps": 78, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0456-hailo-8-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 18, "latency_ms": 55.56, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0285-hailo-8-yolov8-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 68.7, "latency_ms": 14.56, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_pose|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0344-hailo-8-yolov8-pose-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 943, "latency_ms": 1.06, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_pose|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0342-hailo-8-yolov8-pose-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 393, "latency_ms": 2.54, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_pose|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0343-hailo-8-yolov8-pose-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 27.6, "latency_ms": 36.23, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_seg|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0340-hailo-8-yolov8-seg-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 51.2, "latency_ms": 19.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0205-hailo-8-yolov8-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 528, "latency_ms": 1.89, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0203-hailo-8-yolov8-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 107, "latency_ms": 9.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0204-hailo-8-yolov8-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "x", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 14.4, "latency_ms": 69.44, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8|yolov8_seg|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0341-hailo-8-yolov8-seg-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "detr", "series": "detr", "variant": "resnet18", "task": "detection" }, "input": { "resolution": "800x800", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 23.4, "latency_ms": 42.74, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|detr|resnet18|hailo_sdk|int8|800x800|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0194-hailo-8l-detr-resnet18-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8l-ssd-mobilenetv1-estimated-int8-109", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "ssd-mobilenet-v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1200, "latency_ms": 6.7, "latency_p99_ms": 8.5, "tokens_per_s": null, "power_w": 1.2, "memory_mb": null, "accuracy_map50_95": 0.215 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record hailo8l-ssd-mobilenetv1-estimated-int8-109", "notes": "SSD-MobileNetV1 at 300x300 is very light (2.3B FLOPs). Hailo-8L at 13 TOPS is well-matched. 1200 FPS at batch=8.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record hailo8l-ssd-mobilenetv1-estimated-int8-109" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|mobilenet|ssd-mobilenet-v1|hailo_sdk|int8|300x300|8", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0541-hailo-8l-mobilenet-ssd-mobilenet-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 2565, "latency_ms": 0.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|mobilenet|v1|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0216-hailo-8l-mobilenet-v1-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8l-mobilenetv2-int8-144", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 550, "latency_ms": 1.8, "latency_p99_ms": 2.3, "tokens_per_s": null, "power_w": 1.5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8L 13 TOPS, batch=8.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8L classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8L_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|mobilenet|v2|hailo_sdk|int8|224x224|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0564-hailo-8l-mobilenet-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 1738, "latency_ms": 0.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · Hailo Model Zoo classification table, mobilenet_v2_1.0: FPS (Batch Size=1) 1738, FPS (Batch Size=8) 1738; top-1 71.8 float, 70.9 on hardware. System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x 4 lanes, room temperature. Hailo Dataflow Compiler v2.19.0.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo classification", "doc_id": "bench:hailo-model-zoo-classification", "revision": "v2.19.0", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_classification.rst", "retrieved": "2026-09-29", "verified": "2026-09-29", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|mobilenetv2|v2|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0613-hailo-8l-mobilenetv2-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 138, "latency_ms": 7.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|resnet|50|hailo_sdk|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0196-hailo-8l-resnet-50-hailo-sdk-int8" }, { "origin": "research_corpus", "record_id": "hailo8l-resnet50-int8-145", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 8 }, "runtime": "hailo_sdk", "runtime_inferred": true, "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 172, "latency_ms": 5.8, "latency_p99_ms": 7.3, "tokens_per_s": null, "power_w": 2.5, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "Hailo-8L 13 TOPS, batch=8.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — HAILO8L classification", "doc_id": "github:hailo-ai/hailo_model_zoo/HAILO8L_classification", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_classification.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|resnet|50|hailo_sdk|int8|224x224|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0565-hailo-8l-resnet-50-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 356, "latency_ms": 2.81, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|ssd_mobilenet|v1|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0191-hailo-8l-ssd-mobilenet-v1-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v2", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 131, "latency_ms": 7.63, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|ssd_mobilenet|v2|hailo_sdk|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0192-hailo-8l-ssd-mobilenet-v2-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 21.8, "latency_ms": 45.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|l|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 17.5, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0062-hailo-8l-yolo11-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 35.3, "latency_ms": 28.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|m|hailo_sdk|int8|640x640|1|m2", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 28, "tokens_per_s": null, "source": "bench:estimated-from-hailo-8-at-50-tops-ratio" }, { "origin": "catalog", "class": "D", "fps": 27, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0052-hailo-8l-yolo11-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 157, "latency_ms": 6.37, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 67, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "catalog", "class": "C", "fps": 157, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0061-hailo-8l-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 371, "latency_ms": 2.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0448-hailo-8l-yolo11-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 92, "latency_ms": 10.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|s|hailo_sdk|int8|640x640|1|m2", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 52, "tokens_per_s": null, "source": "bench:estimated-from-hailo-8-at-50-tops-ratio" }, { "origin": "catalog", "class": "D", "fps": 44, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "catalog", "class": "D", "fps": 91.9, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0051-hailo-8l-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 192, "latency_ms": 5.21, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0449-hailo-8l-yolo11-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 12.1, "latency_ms": 82.64, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0274-hailo-8l-yolo11-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 34, "latency_ms": 29.41, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_pose|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0385-hailo-8l-yolo11-pose-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 151, "latency_ms": 6.62, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_pose|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0383-hailo-8l-yolo11-pose-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 88.6, "latency_ms": 11.29, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_pose|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0384-hailo-8l-yolo11-pose-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 28.9, "latency_ms": 34.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0382-hailo-8l-yolo11-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 129, "latency_ms": 7.75, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0380-hailo-8l-yolo11-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 75.3, "latency_ms": 13.28, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolo11_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0381-hailo-8l-yolo11-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 150, "latency_ms": 6.67, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov10|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0451-hailo-8l-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 359, "latency_ms": 2.79, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov10|n|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0452-hailo-8l-yolov10-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 87.6, "latency_ms": 11.42, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov10|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0453-hailo-8l-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 187, "latency_ms": 5.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov10|s|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0454-hailo-8l-yolov10-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 60.4, "latency_ms": 16.56, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0200-hailo-8l-yolov5-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 124, "latency_ms": 8.06, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 124, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0199-hailo-8l-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 243, "latency_ms": 4.12, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5|s|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 68, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8L_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0450-hailo-8l-yolov5-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 53, "latency_ms": 18.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0214-hailo-8l-yolov5-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 145, "latency_ms": 6.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0212-hailo-8l-yolov5-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov5_seg", "series": "yolov5_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 102, "latency_ms": 9.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov5_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0213-hailo-8l-yolov5-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 26.1, "latency_ms": 38.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|l|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0279-hailo-8l-yolov8-l-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 51, "latency_ms": 19.61, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|m|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 50.8, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0278-hailo-8l-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 87, "latency_ms": 11.49, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|m|hailo_sdk|int8|640x640|8", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0447-hailo-8l-yolov8-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 202, "latency_ms": 4.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|n|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 202, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" }, { "origin": "research_corpus", "class": "C", "fps": 65, "tokens_per_s": null, "source": "github:hailo-ai/hailo-rpi5-examples" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0276-hailo-8l-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 438, "latency_ms": 2.28, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|n|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 190, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "research_corpus", "class": "C", "fps": 53, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8L_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0445-hailo-8l-yolov8-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 110, "latency_ms": 9.09, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|s|hailo_sdk|int8|640x640|1", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 110, "tokens_per_s": null, "source": "bench:hailo-hailo-model-zoo-object-detection-benchmarks" }, { "origin": "research_corpus", "class": "C", "fps": 42, "tokens_per_s": null, "source": "github:hailo-ai/hailo-rpi5-examples" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=1)" }, "id": "b0277-hailo-8l-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": { "name": "hailo_dataflow_compiler", "version": "v2.19.0" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 8 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 208, "latency_ms": 4.81, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only · System host Intel Core i5-9400 @ 2.90GHz; PCIe Gen 3 x4 lanes, room temperature. FPS column header verbatim: FPS (Batch Size=1) or FPS (Batch Size=8). HEF INT8 via Hailo Dataflow Compiler; zoo table has no per-row precision column.", "sources": [ { "publisher": "Hailo", "document": "Hailo Model Zoo — object detection benchmarks", "doc_id": "bench:hailo-hailo-model-zoo-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8L/HAILO8L_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|s|hailo_sdk|int8|640x640|8", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 38, "tokens_per_s": null, "source": "github:hailo-ai/hailo_model_zoo/HAILO8L_object_detection" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "measurement_distinguisher": { "field": "zoo_fps_column", "value": "FPS (Batch Size=8)" }, "id": "b0446-hailo-8l-yolov8-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 16.2, "latency_ms": 61.73, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8|x|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0280-hailo-8l-yolov8-x-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 48.5, "latency_ms": 20.62, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_pose|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0379-hailo-8l-yolov8-pose-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 195, "latency_ms": 5.13, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_pose|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0377-hailo-8l-yolov8-pose-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 107, "latency_ms": 9.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": "hw_only", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_pose|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0378-hailo-8l-yolov8-pose-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 42, "latency_ms": 23.81, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_seg|m|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0208-hailo-8l-yolov8-seg-m-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 168, "latency_ms": 5.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_seg|n|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0206-hailo-8l-yolov8-seg-n-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "hailo_8l", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "hailo_sdk", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 87.8, "latency_ms": 11.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "Hailo Model Zoo — HAILO8 Object Detection Benchmarks", "doc_id": "bench:publisher-hailo-model-zoo-hailo8-object-detection-benchmarks", "url": "https://github.com/hailo-ai/hailo_model_zoo/blob/master/docs/public_models/HAILO8/HAILO8_object_detection.rst", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "hailo_8l|yolov8_seg|s|hailo_sdk|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0207-hailo-8l-yolov8-seg-s-hailo-sdk-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 310, "latency_ms": 3.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|efficientdet|d0|tensorrt|fp16|512x512|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0124-jetson-agx-orin-efficientdet-d0-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 395.8, "latency_ms": 2.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|efficientdet|d0|tensorrt|int8|512x512|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0125-jetson-agx-orin-efficientdet-d0-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d1", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 198, "latency_ms": 5.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|efficientdet|d1|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0127-jetson-agx-orin-efficientdet-d1-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d2", "task": "detection" }, "input": { "resolution": "768x768", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 98, "latency_ms": 10.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|efficientdet|d2|tensorrt|int8|768x768|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0128-jetson-agx-orin-efficientdet-d2-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "efficientnet-b0-agx-orin-fp16-104", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "efficientnet", "series": "efficientnet", "variant": "b0", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2800, "latency_ms": 0.36, "latency_p99_ms": 0.5, "tokens_per_s": null, "power_w": null, "memory_mb": 220, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record efficientnet-b0-agx-orin-fp16-104", "notes": "EfficientNet-B0 0.39B FLOPs, 5.3M params. ImageNet Top-1 ~77.1%. Jetson AGX Orin GPU far faster than Coral EdgeTPU for FP16.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record efficientnet-b0-agx-orin-fp16-104" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|efficientnet|b0|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0539-jetson-agx-orin-efficientnet-b0-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "efficientnet-b4-agx-orin-fp16-105", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "efficientnet", "series": "efficientnet", "variant": "b4", "task": "classification" }, "input": { "resolution": "380x380", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 680, "latency_ms": 1.47, "latency_p99_ms": 2, "tokens_per_s": null, "power_w": null, "memory_mb": 580, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record efficientnet-b4-agx-orin-fp16-105", "notes": "EfficientNet-B4 4.2B FLOPs, 19M params. ImageNet Top-1 ~82.9%. Non-linear scaling due to larger input.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record efficientnet-b4-agx-orin-fp16-105" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|efficientnet|b4|tensorrt|fp16|380x380|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0540-jetson-agx-orin-efficientnet-b4-tensorrt-fp16" }, { "origin": "p185", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.2-1b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "mlc", "precision": "int4", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 163.9, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.2 1B Instruct (INT4 (q4f16_ft))", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "tutorial_slm.html", "doc_id": "jetson-ai-lab:tutorial_slm.html", "url": "https://www.jetson-ai-lab.com/archive/tutorial_slm.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama-3.2-1B Jetson Orin Nano 54.8 tokens/sec, Jetson AGX Orin 163.9 tokens/sec" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|llama|llama-3.2-1b-instruct|mlc|int4||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0581-jetson-agx-orin-llama-llama-3-2-1b-instruct-mlc-int4" }, { "origin": "p185", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.2-3b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "mlc", "precision": "int4", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 80.4, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.2 3B Instruct (INT4 (q4f16_ft))", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "tutorial_slm.html", "doc_id": "jetson-ai-lab:tutorial_slm.html", "url": "https://www.jetson-ai-lab.com/archive/tutorial_slm.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama-3.2-3B Jetson Orin Nano 27.7 tokens/sec, Jetson AGX Orin 80.4 tokens/sec" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|llama|llama-3.2-3b-instruct|mlc|int4||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0583-jetson-agx-orin-llama-llama-3-2-3b-instruct-mlc-int4" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "mobilenet_ssd", "series": "mobilenet_ssd", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 512, "latency_ms": 1.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|mobilenet_ssd|v1|tensorrt|int8|300x300|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0136-jetson-agx-orin-mobilenet-ssd-v1-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "mobilesam-agx-orin-fp32-129", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "mobilesam", "series": "mobilesam", "variant": "tiny-vit", "task": "segmentation" }, "input": { "resolution": "1024x1024", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 25.6, "latency_ms": 39, "latency_p99_ms": 48, "tokens_per_s": null, "power_w": 30, "memory_mb": 3500, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "MobileSAM full pipeline 39ms on AGX Orin. Image encoder alone: 35ms. mIoU=0.728.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "nanosam README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/nanosam", "url": "https://github.com/NVIDIA-AI-IOT/nanosam", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|mobilesam|tiny-vit|tensorrt|fp32|1024x1024|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0556-jetson-agx-orin-mobilesam-tiny-vit-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "nanosam-agx-orin-fp16-127", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "nanosam", "series": "nanosam", "variant": "resnet18", "task": "segmentation" }, "input": { "resolution": "1024x1024", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 123.5, "latency_ms": 8.1, "latency_p99_ms": 10, "tokens_per_s": null, "power_w": 25, "memory_mb": 1800, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "NanoSAM full pipeline 8.1ms on AGX Orin. Image encoder alone: 4.2ms. mIoU=0.706.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "nanosam README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/nanosam", "url": "https://github.com/NVIDIA-AI-IOT/nanosam", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|nanosam|resnet18|tensorrt|fp16|1024x1024|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0554-jetson-agx-orin-nanosam-resnet18-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "pointpillars-agx-orin-estimated-fp16-097", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "pointpillars", "series": "pointpillars", "variant": "base", "task": "lidar_detection_3d" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 25, "latency_ms": 40, "latency_p99_ms": 55, "tokens_per_s": null, "power_w": 25, "memory_mb": 2800, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record pointpillars-agx-orin-estimated-fp16-097", "notes": "PointPillars KITTI dataset, 3D detection. Community benchmarks report 20-30 FPS on Jetson AGX Orin at FP16 TensorRT. KITTI AP not normalized to COCO.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record pointpillars-agx-orin-estimated-fp16-097" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|pointpillars|base|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0534-jetson-agx-orin-pointpillars-base-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "resnet101-agx-orin-fp16-102", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "101", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1200, "latency_ms": 0.83, "latency_p99_ms": 1.1, "tokens_per_s": null, "power_w": null, "memory_mb": 840, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record resnet101-agx-orin-fp16-102", "notes": "ResNet-101 ~7.9B FLOPs vs ResNet-50 ~4.1B FLOPs. Linear scaling approximation for GPU bound workloads. ImageNet Top-1 ~77.4%.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record resnet101-agx-orin-fp16-102" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|resnet|101|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0538-jetson-agx-orin-resnet-101-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2800, "latency_ms": 0.36, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|resnet|50|tensorrt|fp16|224x224|1|maxn", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 2100, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0187-jetson-agx-orin-resnet-50-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "resnet50-agx-orin-mlperf-int8-077", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1566, "latency_ms": 0.64, "latency_p99_ms": 0.85, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "MLPerf offline scenario, 6423.63 samples/sec, latency 0.64ms, ImageNet", "sources": [ { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|resnet|50|tensorrt|int8|224x224|1|maxn", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 4200, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" }, { "origin": "research_corpus", "class": "C", "fps": 6423.6, "tokens_per_s": null, "source": "mlcommons:inference-edge" } ], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0188-jetson-agx-orin-resnet-50-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "retinanet-agx-orin-mlperf-int8-078", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "retinanet", "series": "retinanet", "variant": "base", "task": "detection" }, "input": { "resolution": "800x800", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 85.7, "latency_ms": 11.67, "latency_p99_ms": 14.5, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "MLPerf offline scenario, 148.71 samples/sec, COCO detection", "sources": [ { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|retinanet|base|tensorrt|int8|800x800|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0519-jetson-agx-orin-retinanet-base-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "mlperf-retinanet-agx-orin-fp16-164", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "retinanet", "series": "retinanet", "variant": "resnet50", "task": "detection" }, "input": { "resolution": "800x800", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 85.7, "latency_ms": 11.67, "latency_p99_ms": 14.5, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q4", "derivation": null, "notes": "MLPerf Inference v3.1 latency scenario RetinaNet on Jetson AGX Orin 64GB. FP16 TensorRT.", "sources": [ { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|retinanet|resnet50|tensorrt|fp16|800x800|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0578-jetson-agx-orin-retinanet-resnet50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 101, "latency_ms": 9.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|rtdetr|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0117-jetson-agx-orin-rtdetr-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 134, "latency_ms": 7.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|rtdetr|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0118-jetson-agx-orin-rtdetr-l-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "whisper-base-agx-orin-fp16-089", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "base", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 560, "latency_p99_ms": 720, "tokens_per_s": null, "power_w": 18, "memory_mb": 560, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record whisper-base-agx-orin-fp16-089", "notes": "Whisper base 74M params, real-time factor ~5.5x on Jetson AGX Orin FP16. WER ~4.2% en. Latency per 10s audio chunk.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record whisper-base-agx-orin-fp16-089" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|whisper|base|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0526-jetson-agx-orin-whisper-base-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-small-agx-orin-fp16-090", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "small", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 1250, "latency_p99_ms": 1600, "tokens_per_s": null, "power_w": 22, "memory_mb": 1100, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record whisper-small-agx-orin-fp16-090", "notes": "Whisper small 244M params, real-time factor ~2.5x on Jetson AGX Orin FP16. WER ~3.0% en. Latency per 10s audio chunk.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record whisper-small-agx-orin-fp16-090" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|whisper|small|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0527-jetson-agx-orin-whisper-small-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-tiny-agx-orin-fp16-088", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "tiny", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 320, "latency_p99_ms": 420, "tokens_per_s": null, "power_w": 15, "memory_mb": 300, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record whisper-tiny-agx-orin-fp16-088", "notes": "Whisper tiny 39M params, ~1GB VRAM, real-time factor ~8x on Jetson AGX Orin FP16 (A100 is ~10x). WER ~5.7% en. Latency per 10s audio chunk.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record whisper-tiny-agx-orin-fp16-088" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|whisper|tiny|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0525-jetson-agx-orin-whisper-tiny-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 124, "latency_ms": 8.08, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 123.8, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0014-jetson-agx-orin-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "pytorch", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 31, "latency_ms": 32.61, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|l|pytorch|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0019-jetson-agx-orin-yolo11-l-pytorch-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 37, "latency_ms": 27, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo11|l|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0108-jetson-agx-orin-yolo11-l-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 163, "latency_ms": 6.12, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|l|tensorrt|int8|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 163.4, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0009-jetson-agx-orin-yolo11-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 160, "latency_ms": 6.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 160, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0013-jetson-agx-orin-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "pytorch", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 40, "latency_ms": 24.88, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|m|pytorch|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0018-jetson-agx-orin-yolo11-m-pytorch-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 60, "latency_ms": 16.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo11|m|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0107-jetson-agx-orin-yolo11-m-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 213, "latency_ms": 4.69, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|m|tensorrt|int8|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 213.2, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0008-jetson-agx-orin-yolo11-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 392, "latency_ms": 2.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 392.2, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0011-jetson-agx-orin-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "pytorch", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 106, "latency_ms": 9.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|n|pytorch|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0016-jetson-agx-orin-yolo11-n-pytorch-fp32" }, { "origin": "research_corpus", "record_id": "yolo11n-agx-orin-fp32-003", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 254.5, "latency_ms": 3.93, "latency_p99_ms": 4.8, "tokens_per_s": null, "power_w": null, "memory_mb": 12100, "accuracy_map50_95": 0.5085 }, "class": "C", "confidence": "high", "date": "2025-Q1", "derivation": null, "notes": "TensorRT FP32, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|yolo11|n|tensorrt|fp32|640x640|1|maxn", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 196, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0105-jetson-agx-orin-yolo11-n-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 459, "latency_ms": 2.18, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|n|tensorrt|int8|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 458.7, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0006-jetson-agx-orin-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 274, "latency_ms": 3.65, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|s|tensorrt|fp16|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 273.9, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0012-jetson-agx-orin-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "pytorch", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 83, "latency_ms": 12.05, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|s|pytorch|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0017-jetson-agx-orin-yolo11-s-pytorch-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 114, "latency_ms": 8.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo11|s|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0106-jetson-agx-orin-yolo11-s-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 356, "latency_ms": 2.81, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|s|tensorrt|int8|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 355.9, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0007-jetson-agx-orin-yolo11-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 74, "latency_ms": 13.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|x|tensorrt|fp16|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 73.8, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0015-jetson-agx-orin-yolo11-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "pytorch", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 17, "latency_ms": 57.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|x|pytorch|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0020-jetson-agx-orin-yolo11-x-pytorch-fp32" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": null, "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 106, "latency_ms": 9.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 60, "source_value": "60w" }, "key": "jetson_agx_orin|yolo11|x|tensorrt|int8|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 106.4, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0010-jetson-agx-orin-yolo11-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": { "name": "tensorrt", "version": "10.3.0" }, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 365.51, "latency_ms": 2.736, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11n-seg FP16 TensorRT model-only benchmark: three rounds of 100 warm-up and 1,000 timed CUDA-synchronized iterations, batch 1, 640x640, mean 2.736 ± 0.031 ms, fps from the mean; MAXN with jetson_clocks; JetPack 6.2.2, TensorRT 10.3.0, CUDA 12.6; Jetson AGX Orin Developer Kit (2048-core GPU, the AGX Orin 64GB compute configuration; developer-kit memory size not stated); model trained on a 6-class cotton-weed dataset (2.836 M parameters, 9.6 GFLOPs)", "sources": [ { "publisher": "Agronomy (MDPI)", "document": "Shan et al., A Lightweight GDMM-YOLO11 Model for Cotton–Weed Instance Segmentation (Agronomy 2026, 16, 1828)", "doc_id": "bench:shan-et-al-a-lightweight-gdmm-yolo11-model-for-cotton-weed-instance-segmentation-agronomy-2026-16-1828", "revision": "Agronomy 2026, 16(18), 1828", "url": "https://doi.org/10.3390/agronomy16181828", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Section 3.7, Table 8 — TensorRT FP16 row", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo11_seg|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0609-jetson-agx-orin-yolo11-seg-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 95, "latency_ms": 10.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0154-jetson-agx-orin-yolo12-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 114, "latency_ms": 8.77, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0155-jetson-agx-orin-yolo12-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 125, "latency_ms": 8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0152-jetson-agx-orin-yolo12-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 150, "latency_ms": 6.67, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0153-jetson-agx-orin-yolo12-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 333, "latency_ms": 3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0148-jetson-agx-orin-yolo12-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 416.5, "latency_ms": 2.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0149-jetson-agx-orin-yolo12-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 232, "latency_ms": 4.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0150-jetson-agx-orin-yolo12-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 283, "latency_ms": 3.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0151-jetson-agx-orin-yolo12-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 51, "latency_ms": 19.61, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0156-jetson-agx-orin-yolo12-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 61.2, "latency_ms": 16.34, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolo12|x|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0157-jetson-agx-orin-yolo12-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 108, "latency_ms": 9.26, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 108, "tokens_per_s": null, "source": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results" }, { "origin": "research_corpus", "class": "D", "fps": 88, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0103-jetson-agx-orin-yolov8-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 163, "latency_ms": 6.13, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0160-jetson-agx-orin-yolov8-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 171, "latency_ms": 5.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 171, "tokens_per_s": null, "source": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results" }, { "origin": "research_corpus", "class": "D", "fps": 122, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0102-jetson-agx-orin-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 248, "latency_ms": 4.03, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0159-jetson-agx-orin-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 371, "latency_ms": 2.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 410, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0099-jetson-agx-orin-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 504, "latency_ms": 1.98, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0100-jetson-agx-orin-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 252, "latency_ms": 3.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 252, "tokens_per_s": null, "source": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results" }, { "origin": "research_corpus", "class": "D", "fps": 245, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0101-jetson-agx-orin-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 342, "latency_ms": 2.92, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0158-jetson-agx-orin-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 67, "latency_ms": 14.93, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [ { "origin": "catalog", "class": "C", "fps": 67, "tokens_per_s": null, "source": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results" }, { "origin": "research_corpus", "class": "D", "fps": 55, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0104-jetson-agx-orin-yolov8-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 107, "latency_ms": 9.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin|yolov8|x|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0161-jetson-agx-orin-yolov8-x-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolov8m-pose-agx-orin-fp16-096", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8", "variant": "m-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 72, "latency_ms": 13.9, "latency_p99_ms": 17, "tokens_per_s": null, "power_w": null, "memory_mb": 3700, "accuracy_map50_95": 0.65 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8m-pose-agx-orin-fp16-096", "notes": "YOLOv8m-pose is ~13% heavier than YOLO11m-pose by FLOPs. Interpolated.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8m-pose-agx-orin-fp16-096" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|yolov8_pose|m-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0533-jetson-agx-orin-yolov8-pose-m-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8m-seg-agx-orin-fp16-094", "hardware": "jetson_agx_orin", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8", "variant": "m-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 95, "latency_ms": 10.5, "latency_p99_ms": 12.8, "tokens_per_s": null, "power_w": null, "memory_mb": 4800, "accuracy_map50_95": 0.499 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8m-seg-agx-orin-fp16-094", "notes": "Interpolated. YOLOv8m-seg 110.2B FLOPs vs YOLO11m-seg 113.2B FLOPs—near equivalent.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8m-seg-agx-orin-fp16-094" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_agx_orin|yolov8_seg|m-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0531-jetson-agx-orin-yolov8-seg-m-seg-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 43.7, "latency_ms": 22.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0082-jetson-agx-orin-32gb-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 73.4, "latency_ms": 13.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0081-jetson-agx-orin-32gb-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 281.6, "latency_ms": 3.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0078-jetson-agx-orin-32gb-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 358.3, "latency_ms": 2.79, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0079-jetson-agx-orin-32gb-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 182.4, "latency_ms": 5.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0080-jetson-agx-orin-32gb-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": { "name": "tensorrt", "version": "10.3.0.30" }, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 322.9, "latency_ms": 3.09, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11s-Pose FP16; trtexec --useCudaGraph --noDataTransfers on an idle device, batch 1, fixed 1x3x640x640, GPU compute mean 3.09429 ms, 322.919 qps; TensorRT 10.3.0.30, CUDA 12.6, L4T R36.4.4; MAXN; AGX Orin 32GB on a reComputer J501 carrier; raw: platforms/jetson/evaluation/orin-multimodule-cuda-stream-fix-20260820.json", "sources": [ { "publisher": "EdgeFallKit", "document": "EdgeFallKit Jetson README — multi-module TensorRT measurements", "doc_id": "bench:edgefallkit-jetson-readme-multi-module-tensorrt-measurements", "revision": "2026-08-20", "url": "https://github.com/suharvest/edgefallkit/blob/main/platforms/jetson/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Process sharding for stream density — trtexec core column", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolo11_pose|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0606-jetson-agx-orin-32gb-yolo11-pose-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 84, "latency_ms": 11.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0180-jetson-agx-orin-32gb-yolov8-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 127, "latency_ms": 7.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0181-jetson-agx-orin-32gb-yolov8-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 130, "latency_ms": 7.69, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0178-jetson-agx-orin-32gb-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 189, "latency_ms": 5.29, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0179-jetson-agx-orin-32gb-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 271, "latency_ms": 3.69, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0174-jetson-agx-orin-32gb-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 352, "latency_ms": 2.84, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0175-jetson-agx-orin-32gb-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 187, "latency_ms": 5.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0176-jetson-agx-orin-32gb-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 254, "latency_ms": 3.94, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0177-jetson-agx-orin-32gb-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 53, "latency_ms": 18.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0182-jetson-agx-orin-32gb-yolov8-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_agx_orin_32gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 85, "latency_ms": 11.76, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_agx_orin_32gb|yolov8|x|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0183-jetson-agx-orin-32gb-yolov8-x-tensorrt-int8" }, { "origin": "p185", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.2-1b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "mlc", "precision": "int4", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 54.8, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.2 1B Instruct (INT4 (q4f16_ft))", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "tutorial_slm.html", "doc_id": "jetson-ai-lab:tutorial_slm.html", "url": "https://www.jetson-ai-lab.com/archive/tutorial_slm.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama-3.2-1B Jetson Orin Nano 54.8 tokens/sec, Jetson AGX Orin 163.9 tokens/sec" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|llama|llama-3.2-1b-instruct|mlc|int4||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0580-jetson-orin-nano-llama-llama-3-2-1b-instruct-mlc-int4" }, { "origin": "p185", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.2-3b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "mlc", "precision": "int4", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 27.7, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.2 3B Instruct (INT4 (q4f16_ft))", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "tutorial_slm.html", "doc_id": "jetson-ai-lab:tutorial_slm.html", "url": "https://www.jetson-ai-lab.com/archive/tutorial_slm.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama-3.2-3B Jetson Orin Nano 27.7 tokens/sec, Jetson AGX Orin 80.4 tokens/sec" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|llama|llama-3.2-3b-instruct|mlc|int4||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0582-jetson-orin-nano-llama-llama-3-2-3b-instruct-mlc-int4" }, { "origin": "research_corpus", "record_id": "mobilenetv2-orin-nano-estimated-fp16-099", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1800, "latency_ms": 0.56, "latency_p99_ms": 0.8, "tokens_per_s": null, "power_w": 5, "memory_mb": 220, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record mobilenetv2-orin-nano-estimated-fp16-099", "notes": "MobileNetV2 ImageNet classifier is lighter than detection heads. Jetson Orin Nano GPU (1024 CUDA cores) gives significant boost over Xavier. ImageNet Top-1 ~71.8%.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record mobilenetv2-orin-nano-estimated-fp16-099" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|mobilenet|v2|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0536-jetson-orin-nano-mobilenet-v2-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "mobilenetv3-orin-nano-estimated-fp16-100", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v3-large", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1600, "latency_ms": 0.63, "latency_p99_ms": 0.9, "tokens_per_s": null, "power_w": 5, "memory_mb": 260, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record mobilenetv3-orin-nano-estimated-fp16-100", "notes": "MobileNetV3-Large slightly heavier than V2. ImageNet Top-1 ~75.2%.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record mobilenetv3-orin-nano-estimated-fp16-100" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|mobilenet|v3-large|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0537-jetson-orin-nano-mobilenet-v3-large-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "mobilesam-orin-nano-fp32-128", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "mobilesam", "series": "mobilesam", "variant": "tiny-vit", "task": "segmentation" }, "input": { "resolution": "1024x1024", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 6.8, "latency_ms": 146, "latency_p99_ms": 180, "tokens_per_s": null, "power_w": 10, "memory_mb": 3500, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "MobileSAM Tiny-ViT encoder + SAM decoder, full pipeline 146ms on Orin Nano. mIoU=0.728 on COCO 2017 val.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "nanosam README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/nanosam", "url": "https://github.com/NVIDIA-AI-IOT/nanosam", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|mobilesam|tiny-vit|tensorrt|fp32|1024x1024|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0555-jetson-orin-nano-mobilesam-tiny-vit-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "nanosam-orin-nano-fp16-126", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "nanosam", "series": "nanosam", "variant": "resnet18", "task": "segmentation" }, "input": { "resolution": "1024x1024", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 37, "latency_ms": 27, "latency_p99_ms": 35, "tokens_per_s": null, "power_w": 9, "memory_mb": 1800, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "NanoSAM (ResNet18 encoder distilled from MobileSAM) + SAM mask decoder, TensorRT FP16, full pipeline 27ms. mIoU=0.706 on COCO 2017 val.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "nanosam README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/nanosam", "url": "https://github.com/NVIDIA-AI-IOT/nanosam", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|nanosam|resnet18|tensorrt|fp16|1024x1024|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0553-jetson-orin-nano-nanosam-resnet18-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-base-orin-nano-trt-fp16-131", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "base-en", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 860, "latency_p99_ms": 1100, "tokens_per_s": null, "power_w": 8.5, "memory_mb": 439, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "WhisperTRT (TensorRT FP16) on Jetson Orin Nano. 20-second audio clip processed in 0.86s = 23.3x real-time. ~3x faster than PyTorch. Memory 439MB (vs 666MB PyTorch).", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "whisper_trt README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/whisper_trt", "url": "https://github.com/NVIDIA-AI-IOT/whisper_trt", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|whisper|base-en|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0558-jetson-orin-nano-whisper-base-en-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-tiny-orin-nano-trt-fp16-130", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "tiny-en", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 640, "latency_p99_ms": 820, "tokens_per_s": null, "power_w": 8, "memory_mb": 488, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "WhisperTRT (TensorRT FP16) on Jetson Orin Nano. 20-second audio clip processed in 0.64s = 31.25x real-time factor. Peak memory 488MB. English-only model.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "whisper_trt README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/whisper_trt", "url": "https://github.com/NVIDIA-AI-IOT/whisper_trt", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|whisper|tiny-en|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0557-jetson-orin-nano-whisper-tiny-en-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-tiny-orin-nano-pytorch-fp32-132", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "tiny-en", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 1740, "latency_p99_ms": 2200, "tokens_per_s": null, "power_w": 8, "memory_mb": 569, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "PyTorch Whisper FP32 baseline on Orin Nano. 20s audio takes 1.74s. Memory 569MB. Comparison baseline for TRT speedup measurement.", "sources": [ { "publisher": "NVIDIA (NVIDIA-AI-IOT)", "document": "whisper_trt README benchmark table", "doc_id": "github:NVIDIA-AI-IOT/whisper_trt", "url": "https://github.com/NVIDIA-AI-IOT/whisper_trt", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|whisper|tiny-en|tensorrt|fp32|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0559-jetson-orin-nano-whisper-tiny-en-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 18.7, "latency_ms": 53.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0077-jetson-orin-nano-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 31.4, "latency_ms": 31.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0076-jetson-orin-nano-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 114.8, "latency_ms": 8.71, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0073-jetson-orin-nano-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 151.8, "latency_ms": 6.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano|yolo11|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0074-jetson-orin-nano-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 75.6, "latency_ms": 13.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano|yolo11|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0075-jetson-orin-nano-yolo11-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8m-orin-nano-estimated-fp16-084", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 48, "latency_ms": 20.8, "latency_p99_ms": 25, "tokens_per_s": null, "power_w": 8.5, "memory_mb": 3800, "accuracy_map50_95": 0.502 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8m-orin-nano-estimated-fp16-084", "notes": "YOLOv8m has ~16% more FLOPs than YOLO11m. Interpolated.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8m-orin-nano-estimated-fp16-084" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0601-jetson-orin-nano-yolov8-m-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-orin-nano-estimated-fp16-082", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 190, "latency_ms": 5.26, "latency_p99_ms": 6.4, "tokens_per_s": null, "power_w": 6.2, "memory_mb": 850, "accuracy_map50_95": 0.373 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-orin-nano-estimated-fp16-082", "notes": "Interpolated. YOLOv8n has ~8% more FLOPs than YOLO11n. Orin Nano 8GB = 40 TOPS vs Orin Nano Super ~67 TOPS effectively.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-orin-nano-estimated-fp16-082" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0599-jetson-orin-nano-yolov8-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-orin-nano-estimated-int8-081", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 245, "latency_ms": 4.08, "latency_p99_ms": 5, "tokens_per_s": null, "power_w": 6.5, "memory_mb": 820, "accuracy_map50_95": 0.373 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-orin-nano-estimated-int8-081", "notes": "YOLOv8n is ~10% heavier than YOLO11n; Orin Nano (non-Super) ~15% slower than Orin Nano Super. Interpolated from verified YOLO11 Orin Nano Super benchmarks.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-orin-nano-estimated-int8-081" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|yolov8|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0598-jetson-orin-nano-yolov8-n-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolov8s-orin-nano-estimated-fp16-083", "hardware": "jetson_orin_nano", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 110, "latency_ms": 9.1, "latency_p99_ms": 11, "tokens_per_s": null, "power_w": 7, "memory_mb": 1800, "accuracy_map50_95": 0.449 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8s-orin-nano-estimated-fp16-083", "notes": "Interpolated from YOLO11 Orin Nano Super numbers with YOLOv8/YOLO11 compute ratio adjustment.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8s-orin-nano-estimated-fp16-083" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano|yolov8|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0600-jetson-orin-nano-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 9, "latency_ms": 111.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0596-jetson-orin-nano-4gb-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 13, "latency_ms": 76.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0597-jetson-orin-nano-4gb-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 37, "latency_ms": 27.03, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0592-jetson-orin-nano-4gb-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 48, "latency_ms": 20.83, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0593-jetson-orin-nano-4gb-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 22, "latency_ms": 45.45, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0594-jetson-orin-nano-4gb-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_4gb", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 30, "latency_ms": 33.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_4gb|yolov8|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0595-jetson-orin-nano-4gb-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "clip", "series": "clip_vit", "variant": "base-patch16", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 161, "latency_ms": 6.21, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "NVIDIA", "document": "Publisher benchmark table", "doc_id": "bench:nvidia-publisher-benchmark-table", "url": "https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|clip|base-patch16|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0418-jetson-orin-nano-super-clip-base-patch16-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "clip", "series": "clip_vit", "variant": "base-patch32", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 314, "latency_ms": 3.18, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "NVIDIA", "document": "Publisher benchmark table", "doc_id": "bench:nvidia-publisher-benchmark-table", "url": "https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|clip|base-patch32|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0417-jetson-orin-nano-super-clip-base-patch32-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "dinov2", "series": "dinov2", "variant": "base-patch14", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 126, "latency_ms": 7.94, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "NVIDIA", "document": "Publisher benchmark table", "doc_id": "bench:nvidia-publisher-benchmark-table", "url": "https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|dinov2|base-patch14|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0419-jetson-orin-nano-super-dinov2-base-patch14-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 162, "latency_ms": 6.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|efficientdet|d0|tensorrt|int8|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0122-jetson-orin-nano-super-efficientdet-d0-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d1", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 78, "latency_ms": 12.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|efficientdet|d1|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0126-jetson-orin-nano-super-efficientdet-d1-tensorrt-int8" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "gemma2", "series": "gemma2", "variant": "gemma-2-2b-it", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 34.97, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Gemma 2 2B IT (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Gemma 2 2B: Jetson Orin Nano 21.5, Jetson Orin Nano Super 34.97 (perf gain 1.63x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|gemma2|gemma-2-2b-it|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0587-jetson-orin-nano-super-gemma2-gemma-2-2b-it-unspecified-unspecified" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.1-8b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 19.14, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.1 8B Instruct (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama 3.1 8B: Jetson Orin Nano 14, Jetson Orin Nano Super 19.14 (perf gain 1.37x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|llama|llama-3.1-8b-instruct|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0585-jetson-orin-nano-super-llama-llama-3-1-8b-instruct-unspecified-unspecified" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "llama", "series": "llama", "variant": "llama-3.2-3b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 43.07, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Llama 3.2 3B Instruct (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Llama 3.2 3B: Jetson Orin Nano 27.7, Jetson Orin Nano Super 43.07 (perf gain 1.55x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|llama|llama-3.2-3b-instruct|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0584-jetson-orin-nano-super-llama-llama-3-2-3b-instruct-unspecified-unspecified" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "llava", "series": "llava", "variant": "llava-1.5-7b-hf", "task": "vision_language" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 0.57, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "LLaVA 1.5 7B (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "LLAVA 1.6 7B: Jetson Orin Nano 0.412, Jetson Orin Nano Super 0.57 (perf gain 1.38x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|llava|llava-1.5-7b-hf|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0589-jetson-orin-nano-super-llava-llava-1-5-7b-hf-unspecified-unspecified" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "mobilenet_ssd", "series": "mobilenet_ssd", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 215, "latency_ms": 4.65, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|mobilenet_ssd|v1|tensorrt|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0135-jetson-orin-nano-super-mobilenet-ssd-v1-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "maxn_super", "software": { "name": "jetpack", "version": "L4T R36.4.7, TensorRT 10.3.0" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1176.5, "latency_ms": 0.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "Jetson Orin Nano 8GB at MAXN_SUPER, JetPack 6 (L4T R36.4.7), TensorRT 10.3.0, CUDA 12.6. ONNX model zoo mobilenetv2-12, engine built with --fp16. trtexec --noDataTransfers mean GPU compute time 0.850 ms at batch 1 (CUDA-event timing, transfers excluded); fps = 1000 / 0.850 ms, one stream back to back. Same table: INT8+FP16 0.781 ms, FP32 1.386 ms; batch 8 FP16 3.925 ms.", "sources": [ { "publisher": "onnxsim", "document": "onnxsim README — ModelOpt INT8 and TensorRT latency, Jetson Orin Nano 8GB on JetPack 6", "doc_id": "bench:onnxsim-readme-modelopt-int8-and-tensorrt-latency-jetson-orin-nano-8gb-on-jetpack-6", "revision": "L4T R36.4.7, TensorRT 10.3.0", "url": "https://github.com/onnxsim/onnxsim/blob/387c626ef1e6f40b838034599d6185d8ff54f7a4/scripts/nvidia/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nano_super|mobilenetv2|v2|tensorrt|fp16|224x224|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0610-jetson-orin-nano-super-mobilenetv2-v2-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "maxn_super", "software": { "name": "jetpack", "version": "L4T R36.4.7, TensorRT 10.3.0" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1280.4, "latency_ms": 0.781, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "Jetson Orin Nano 8GB at MAXN_SUPER, JetPack 6 (L4T R36.4.7), TensorRT 10.3.0, CUDA 12.6. ONNX model zoo mobilenetv2-12 quantized by ModelOpt (explicit Q/DQ, entropy calibration on 128 Imagenette train images), engine built with --int8 --fp16. trtexec --noDataTransfers mean GPU compute time 0.781 ms at batch 1 (CUDA-event timing, transfers excluded); fps = 1000 / 0.781 ms, one stream back to back. Same table: FP16 0.850 ms, FP32 1.386 ms; batch 8 INT8 2.446 ms.", "sources": [ { "publisher": "onnxsim", "document": "onnxsim README — ModelOpt INT8 and TensorRT latency, Jetson Orin Nano 8GB on JetPack 6", "doc_id": "bench:onnxsim-readme-modelopt-int8-and-tensorrt-latency-jetson-orin-nano-8gb-on-jetpack-6", "revision": "L4T R36.4.7, TensorRT 10.3.0", "url": "https://github.com/onnxsim/onnxsim/blob/387c626ef1e6f40b838034599d6185d8ff54f7a4/scripts/nvidia/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nano_super|mobilenetv2|v2|tensorrt|int8|224x224|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0611-jetson-orin-nano-super-mobilenetv2-v2-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "mobilenetv3", "series": "mobilenetv3", "variant": "large", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 820, "latency_ms": 1.22, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|mobilenetv3|large|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0133-jetson-orin-nano-super-mobilenetv3-large-tensorrt-fp16" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "qwen2.5", "series": "qwen2.5", "variant": "qwen2.5-7b-instruct", "task": "text_generation" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 21.75, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "Qwen2.5 7B Instruct (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "Qwen2.5 7B: Jetson Orin Nano 14.2, Jetson Orin Nano Super 21.75 (perf gain 1.53x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|qwen2.5|qwen2.5-7b-instruct|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0586-jetson-orin-nano-super-qwen2-5-qwen2-5-7b-instruct-unspecified-unspecified" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 650, "latency_ms": 1.54, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|resnet|50|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0184-jetson-orin-nano-super-resnet-50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 19.8, "latency_ms": 50.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|rtdetr|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0113-jetson-orin-nano-super-rtdetr-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 28.2, "latency_ms": 35.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|rtdetr|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0114-jetson-orin-nano-super-rtdetr-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 12.5, "latency_ms": 80, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|rtdetr|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0119-jetson-orin-nano-super-rtdetr-x-tensorrt-fp16" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "smolvlm", "series": "smolvlm", "variant": "smolvlm-2.2b-instruct", "task": "vision_language" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 12.9, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "SmolVLM 2.2B Instruct (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "SmolVLM 2B: Jetson Orin Nano 8.1, Jetson Orin Nano Super 12.9 (perf gain 1.59x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|smolvlm|smolvlm-2.2b-instruct|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0588-jetson-orin-nano-super-smolvlm-smolvlm-2-2b-instruct-unspecified-unspecified" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "vila", "series": "vila", "variant": "vila-1.5-3b", "task": "vision_language" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 1.06, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "VILA 1.5 3B (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "VILA 1.5 3B: Jetson Orin Nano 0.7, Jetson Orin Nano Super 1.06 (perf gain 1.51x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|vila|vila-1.5-3b|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0590-jetson-orin-nano-super-vila-vila-1-5-3b-unspecified-unspecified" }, { "origin": "p185", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "vila", "series": "vila", "variant": "vila-1.5-8b", "task": "vision_language" }, "input": { "resolution": null, "batch": 1 }, "runtime": "unspecified", "precision": "unspecified", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": null, "latency_p99_ms": null, "tokens_per_s": 0.83, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026-09-07", "derivation": null, "notes": "VILA 1.5 8B (Llama-3) (quant unspecified)", "sources": [ { "publisher": "NVIDIA Jetson AI Lab (archive)", "document": "benchmarks.html", "doc_id": "jetson-ai-lab:benchmarks.html", "url": "https://www.jetson-ai-lab.com/archive/benchmarks.html", "retrieved": "2026-09-07", "verified": "2026-09-07", "class": "C", "quote": "VILA 1.5 8B: Jetson Orin Nano 0.574, Jetson Orin Nano Super 0.83 (perf gain 1.45x)" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|vila|vila-1.5-8b|unspecified|unspecified||1", "duplicates": [], "latency_source": null, "latency_p99_source": null, "latency_p99_basis": null, "id": "b0591-jetson-orin-nano-super-vila-vila-1-5-8b-unspecified-unspecified" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "vit", "series": "vit", "variant": "base-patch16", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 158, "latency_ms": 6.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "NVIDIA", "document": "Publisher benchmark table", "doc_id": "bench:nvidia-publisher-benchmark-table", "url": "https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|vit|base-patch16|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0415-jetson-orin-nano-super-vit-base-patch16-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "vit", "series": "vit", "variant": "base-patch32", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 273, "latency_ms": 3.66, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "NVIDIA", "document": "Publisher benchmark table", "doc_id": "bench:nvidia-publisher-benchmark-table", "url": "https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|vit|base-patch32|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0416-jetson-orin-nano-super-vit-base-patch32-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 57, "latency_ms": 17.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|l|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 57.2, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0029-jetson-orin-nano-super-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 28, "latency_ms": 35.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|l|tensorrt|fp32|640x640|1|15w", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 8, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0034-jetson-orin-nano-super-yolo11-l-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 84, "latency_ms": 11.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|l|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 84.2, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0024-jetson-orin-nano-super-yolo11-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 74, "latency_ms": 13.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|m|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 74.1, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0028-jetson-orin-nano-super-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 36, "latency_ms": 27.86, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|m|tensorrt|fp32|640x640|1|15w", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 14, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0033-jetson-orin-nano-super-yolo11-m-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 110, "latency_ms": 9.12, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|m|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 109.6, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0023-jetson-orin-nano-super-yolo11-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 221, "latency_ms": 4.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|n|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 220.8, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0026-jetson-orin-nano-super-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 134, "latency_ms": 7.44, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|n|tensorrt|fp32|640x640|1|15w", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 52, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0031-jetson-orin-nano-super-yolo11-n-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 270, "latency_ms": 3.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|n|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 270.3, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0021-jetson-orin-nano-super-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 144, "latency_ms": 6.93, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|s|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 144.3, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0027-jetson-orin-nano-super-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 77, "latency_ms": 13.02, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|s|tensorrt|fp32|640x640|1|15w", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 31, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0032-jetson-orin-nano-super-yolo11-s-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 197, "latency_ms": 5.08, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|s|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 196.9, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0022-jetson-orin-nano-super-yolo11-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 31, "latency_ms": 32.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|x|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 30.7, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0030-jetson-orin-nano-super-yolo11-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 15, "latency_ms": 66.93, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|x|tensorrt|fp32|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0035-jetson-orin-nano-super-yolo11-x-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.157" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 50, "latency_ms": 19.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://github.com/ultralytics/ultralytics/blob/v8.3.203/docs/en/guides/nvidia-jetson.md", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nano_super|yolo11|x|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 50.3, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0025-jetson-orin-nano-super-yolo11-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "maxn_super", "software": { "name": "tensorrt", "version": "10.3" }, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 270.7, "latency_ms": 3.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "YOLO11n-Pose FP16; trtexec --useCudaGraph --noDataTransfers, one execution context, batch 1, fixed 1x3x640x640, GPU compute only (no decode, host copies or post-processing); co-resident GPU workloads stopped; MAXN_SUPER; TensorRT 10.3, CUDA 12.6; reComputer J3011 (Jetson Orin Nano 8GB) in Super mode; raw report (EdgeFallKit yolo11n-crossplatform-20260814.json): 270.727 fps, 3.69129 ms at one context", "sources": [ { "publisher": "Seeed Studio", "document": "Seeed Studio wiki — On-Device AI Fall Detection: Build, Deploy, and Measured Results", "doc_id": "bench:seeed-studio-wiki-on-device-ai-fall-detection-build-deploy-and-measured-results", "revision": "2026-09-01", "url": "https://wiki.seeedstudio.com/solutions/edge-fall-detection/", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Detailed performance results — Per-frame latency and throughput (FP16 table)", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nano_super|yolo11_pose|n|tensorrt|fp16|640x640|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0602-jetson-orin-nano-super-yolo11-pose-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": "maxn_super", "software": { "name": "tensorrt", "version": "10.3.0.30" }, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 155.3, "latency_ms": 6.44, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11s-Pose FP16; trtexec --useCudaGraph --noDataTransfers on an idle device, batch 1, fixed 1x3x640x640, GPU compute mean 6.4372 ms, 155.286 qps; TensorRT 10.3.0.30, CUDA 12.6, L4T R36.4.3; MAXN_SUPER; Orin Nano Super 8GB; raw: platforms/jetson/evaluation/orin-multimodule-cuda-stream-fix-20260820.json", "sources": [ { "publisher": "EdgeFallKit", "document": "EdgeFallKit Jetson README — multi-module TensorRT measurements", "doc_id": "bench:edgefallkit-jetson-readme-multi-module-tensorrt-measurements", "revision": "2026-08-20", "url": "https://github.com/suharvest/edgefallkit/blob/main/platforms/jetson/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Process sharding for stream density — trtexec core column", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nano_super|yolo11_pose|s|tensorrt|fp16|640x640|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0604-jetson-orin-nano-super-yolo11-pose-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "tensorrt", "version": "10.3.0" }, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 186.1, "latency_ms": 5.37, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11n-seg FP16 TensorRT engine run by the Ultralytics validator on the device, batch 1, imgsz 640, COCO val2017 (4,999 images); inference stage mean 5.3735 ms per image, fps = 1000 / ms; preprocess (1.61 ms) and postprocess (43.09 ms) are reported apart and not included; TensorRT 10.3.0 / JetPack 6, CUDA 12.6; Jetson Orin Nano Super developer kit (7.4 GiB); power mode not stated", "sources": [ { "publisher": "EdgeFirst AI", "document": "EdgeFirst yolo-validator benchmarks — Jetson Orin Nano Super TensorRT metrics", "doc_id": "bench:edgefirst-yolo-validator-benchmarks-jetson-orin-nano-super-tensorrt-metrics", "revision": "2026-06-22", "url": "https://github.com/EdgeFirstAI/yolo-validator/blob/main/benchmarks/metrics/orin-nano-tensorrt.json", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Ultralytics validator lane, latency_ms.inf", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|yolo11_seg|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0607-jetson-orin-nano-super-yolo11-seg-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "tensorrt", "version": "10.3.0" }, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 120.1, "latency_ms": 8.32, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11s-seg FP16 TensorRT engine run by the Ultralytics validator on the device, batch 1, imgsz 640, COCO val2017 (4,998 images); inference stage mean 8.3241 ms per image, fps = 1000 / ms; preprocess (1.65 ms) and postprocess (36.45 ms) are reported apart and not included; TensorRT 10.3.0 / JetPack 6, CUDA 12.6; Jetson Orin Nano Super developer kit (7.4 GiB); power mode not stated", "sources": [ { "publisher": "EdgeFirst AI", "document": "EdgeFirst yolo-validator benchmarks — Jetson Orin Nano Super TensorRT metrics", "doc_id": "bench:edgefirst-yolo-validator-benchmarks-jetson-orin-nano-super-tensorrt-metrics", "revision": "2026-06-22", "url": "https://github.com/EdgeFirstAI/yolo-validator/blob/main/benchmarks/metrics/orin-nano-tensorrt.json", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Ultralytics validator lane, latency_ms.inf", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|yolo11_seg|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0608-jetson-orin-nano-super-yolo11-seg-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 42.9, "latency_ms": 23.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0144-jetson-orin-nano-super-yolo12-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 51.5, "latency_ms": 19.42, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0145-jetson-orin-nano-super-yolo12-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 57.8, "latency_ms": 17.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0142-jetson-orin-nano-super-yolo12-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 69.4, "latency_ms": 14.42, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0143-jetson-orin-nano-super-yolo12-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 187, "latency_ms": 5.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0138-jetson-orin-nano-super-yolo12-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 228.1, "latency_ms": 4.38, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0139-jetson-orin-nano-super-yolo12-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 118, "latency_ms": 8.47, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0140-jetson-orin-nano-super-yolo12-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 143.9, "latency_ms": 6.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0141-jetson-orin-nano-super-yolo12-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 21.5, "latency_ms": 46.51, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0146-jetson-orin-nano-super-yolo12-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 25.8, "latency_ms": 38.76, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolo12|x|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0147-jetson-orin-nano-super-yolo12-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 41, "latency_ms": 24.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0094-jetson-orin-nano-super-yolov8-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 73, "latency_ms": 13.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0090-jetson-orin-nano-super-yolov8-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 70, "latency_ms": 14.29, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0087-jetson-orin-nano-super-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 108, "latency_ms": 9.26, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0088-jetson-orin-nano-super-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 259, "latency_ms": 3.86, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0083-jetson-orin-nano-super-yolov8-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-orin-nano-super-onnx-fp32-110", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 69.1, "latency_ms": 14.5, "latency_p99_ms": 17.8, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.373 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-orin-nano-super-onnx-fp32-110", "notes": "ONNX runtime FP32 without TensorRT optimization. Reference for integration testing.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-orin-nano-super-onnx-fp32-110" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nano_super|yolov8|n|tensorrt|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0542-jetson-orin-nano-super-yolov8-n-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 341, "latency_ms": 2.93, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0084-jetson-orin-nano-super-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 164, "latency_ms": 6.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0085-jetson-orin-nano-super-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 233, "latency_ms": 4.29, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0086-jetson-orin-nano-super-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 23, "latency_ms": 43.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0091-jetson-orin-nano-super-yolov8-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nano_super", "power_mode": null, "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 44, "latency_ms": 22.73, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nano_super|yolov8|x|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0092-jetson-orin-nano-super-yolov8-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 248, "latency_ms": 4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|efficientdet|d0|tensorrt|int8|512x512|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0123-jetson-orin-nx-efficientdet-d0-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1250, "latency_ms": 0.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|mobilenetv2|v2|tensorrt|fp16|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0131-jetson-orin-nx-mobilenetv2-v2-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "mobilenetv3", "series": "mobilenetv3", "variant": "large", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1420, "latency_ms": 0.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|mobilenetv3|large|tensorrt|fp16|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0134-jetson-orin-nx-mobilenetv3-large-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "pointpillars-orin-nx-estimated-fp16-098", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "pointpillars", "series": "pointpillars", "variant": "base", "task": "lidar_detection_3d" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 14, "latency_ms": 71.4, "latency_p99_ms": 90, "tokens_per_s": null, "power_w": 15, "memory_mb": 2800, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record pointpillars-orin-nx-estimated-fp16-098", "notes": "Orin NX 16GB GPU ~1.8x slower than AGX Orin for LiDAR workloads. KITTI 3D detection.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record pointpillars-orin-nx-estimated-fp16-098" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|pointpillars|base|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0535-jetson-orin-nx-pointpillars-base-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1180, "latency_ms": 0.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|resnet|50|tensorrt|fp16|224x224|1|maxn", "duplicates": [ { "origin": "research_corpus", "class": "D", "fps": 1200, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0185-jetson-orin-nx-resnet-50-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "mlperf-resnet50-orin-nx-int8-163", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1681.9, "latency_ms": 0.595, "latency_p99_ms": 0.75, "tokens_per_s": null, "power_w": 14.95, "memory_mb": 800, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q4", "derivation": null, "notes": "MLPerf Inference v3.1 offline scenario on Jetson Orin NX 16GB. 14.95W measured system power.", "sources": [ { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|resnet|50|tensorrt|int8|224x224|1|maxn", "duplicates": [ { "origin": "catalog", "class": "D", "fps": 1820, "tokens_per_s": null, "source": "edgeaistack:benchmark-corpus" } ], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0186-jetson-orin-nx-resnet-50-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "mlperf-retinanet-orin-nx-fp16-165", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "retinanet", "series": "retinanet", "variant": "resnet50", "task": "detection" }, "input": { "resolution": "800x800", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 21, "latency_ms": 47.59, "latency_p99_ms": 59, "tokens_per_s": null, "power_w": 15.57, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2023-Q4", "derivation": null, "notes": "MLPerf Inference v3.1 latency scenario on Jetson Orin NX 16GB. 15.57W measured.", "sources": [ { "publisher": "MLCommons", "document": "MLPerf Inference: Edge results", "doc_id": "mlcommons:inference-edge", "url": "https://mlcommons.org/benchmarks/inference-edge/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|retinanet|resnet50|tensorrt|fp16|800x800|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0579-jetson-orin-nx-retinanet-resnet50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 46.5, "latency_ms": 21.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|rtdetr|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0115-jetson-orin-nx-rtdetr-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 61.9, "latency_ms": 16.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|rtdetr|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0116-jetson-orin-nx-rtdetr-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 29.8, "latency_ms": 33.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|rtdetr|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0120-jetson-orin-nx-rtdetr-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "rtdetr", "series": "rtdetr", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 40.1, "latency_ms": 24.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|rtdetr|x|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0121-jetson-orin-nx-rtdetr-x-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "whisper-base-orin-nx-fp16-092", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "base", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 950, "latency_p99_ms": 1200, "tokens_per_s": null, "power_w": 12, "memory_mb": 560, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record whisper-base-orin-nx-fp16-092", "notes": "Real-time factor ~3.2x on Orin NX 16GB. Per 10s audio chunk.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record whisper-base-orin-nx-fp16-092" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|whisper|base|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0529-jetson-orin-nx-whisper-base-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "whisper-tiny-orin-nx-fp16-091", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "whisper", "series": "whisper", "variant": "tiny", "task": "speech_recognition" }, "input": { "resolution": "n/a", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": null, "latency_ms": 550, "latency_p99_ms": 720, "tokens_per_s": null, "power_w": 10, "memory_mb": 300, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record whisper-tiny-orin-nx-fp16-091", "notes": "Orin NX 16GB has 1024 CUDA cores vs 2048 on AGX Orin. Real-time factor ~4.5x. Per 10s audio chunk.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record whisper-tiny-orin-nx-fp16-091" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|whisper|tiny|tensorrt|fp16|n/a|1", "duplicates": [], "latency_source": "measured", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0528-jetson-orin-nx-whisper-tiny-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 54, "latency_ms": 18.52, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|l|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 53.8, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0044-jetson-orin-nx-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 16, "latency_ms": 62.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolo11|l|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0112-jetson-orin-nx-yolo11-l-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 79, "latency_ms": 12.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|l|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 79.4, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0039-jetson-orin-nx-yolo11-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 69, "latency_ms": 14.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|m|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 69.1, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0043-jetson-orin-nx-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 27, "latency_ms": 37, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolo11|m|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0111-jetson-orin-nx-yolo11-m-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 103, "latency_ms": 9.74, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|m|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 102.7, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0038-jetson-orin-nx-yolo11-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 211, "latency_ms": 4.73, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|n|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 211.4, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0041-jetson-orin-nx-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 95, "latency_ms": 10.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Normalized research table", "doc_id": "bench:publisher-normalized-research-table", "url": "https://github.com/ultralytics/ultralytics/discussions/14827", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolo11|n|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0109-jetson-orin-nx-yolo11-n-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 256, "latency_ms": 3.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|n|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 256.4, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0036-jetson-orin-nx-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 137, "latency_ms": 7.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|s|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 136.8, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0042-jetson-orin-nx-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 56, "latency_ms": 17.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Normalized research table", "doc_id": "bench:publisher-normalized-research-table", "url": "https://github.com/ultralytics/ultralytics/discussions/14827", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolo11|s|tensorrt|fp32|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0110-jetson-orin-nx-yolo11-s-tensorrt-fp32" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 188, "latency_ms": 5.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|s|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 187.6, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0037-jetson-orin-nx-yolo11-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 28, "latency_ms": 35.71, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|x|tensorrt|fp16|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 27.9, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0045-jetson-orin-nx-yolo11-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": { "name": "ultralytics", "version": "8.3.x" }, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 46, "latency_ms": 21.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2024", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo11|x|tensorrt|int8|640x640|1|15w", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 46.3, "tokens_per_s": null, "source": "ultralytics:guides/nvidia-jetson" } ], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0040-jetson-orin-nx-yolo11-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn_super", "software": { "name": "tensorrt", "version": "10.3" }, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 306.2, "latency_ms": 3.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": "YOLO11n-Pose FP16; trtexec --useCudaGraph --noDataTransfers, one execution context, batch 1, fixed 1x3x640x640, GPU compute only (no decode, host copies or post-processing); co-resident GPU workloads stopped; MAXN_SUPER; TensorRT 10.3, CUDA 12.6; reComputer J4012 (Jetson Orin NX 16GB) in Super mode; raw report (EdgeFallKit yolo11n-crossplatform-20260814.json): 306.154 fps, 3.26376 ms at one context", "sources": [ { "publisher": "Seeed Studio", "document": "Seeed Studio wiki — On-Device AI Fall Detection: Build, Deploy, and Measured Results", "doc_id": "bench:seeed-studio-wiki-on-device-ai-fall-detection-build-deploy-and-measured-results", "revision": "2026-09-01", "url": "https://wiki.seeedstudio.com/solutions/edge-fall-detection/", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Detailed performance results — Per-frame latency and throughput (FP16 table)", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nx|yolo11_pose|n|tensorrt|fp16|640x640|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0603-jetson-orin-nx-yolo11-pose-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn_super", "software": { "name": "tensorrt", "version": "10.3.0.30" }, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 173.8, "latency_ms": 5.75, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "YOLO11s-Pose FP16; trtexec --useCudaGraph --noDataTransfers on an idle device, batch 1, fixed 1x3x640x640, GPU compute mean 5.75119 ms, 173.807 qps; TensorRT 10.3.0.30, CUDA 12.6, L4T R36.4.3; MAXN_SUPER; Orin NX 16GB (Super mode); raw: platforms/jetson/evaluation/orin-multimodule-cuda-stream-fix-20260820.json", "sources": [ { "publisher": "EdgeFallKit", "document": "EdgeFallKit Jetson README — multi-module TensorRT measurements", "doc_id": "bench:edgefallkit-jetson-readme-multi-module-tensorrt-measurements", "revision": "2026-08-20", "url": "https://github.com/suharvest/edgefallkit/blob/main/platforms/jetson/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "section": "Process sharding for stream density — trtexec core column", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn_super" }, "key": "jetson_orin_nx|yolo11_pose|s|tensorrt|fp16|640x640|1|maxn_super", "duplicates": [], "latency_source": "measured", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0605-jetson-orin-nx-yolo11-pose-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 40.5, "latency_ms": 24.69, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|l|tensorrt|fp16|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0408-jetson-orin-nx-yolo12-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 48.6, "latency_ms": 20.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|l|tensorrt|int8|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0413-jetson-orin-nx-yolo12-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 53.8, "latency_ms": 18.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|m|tensorrt|fp16|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0407-jetson-orin-nx-yolo12-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 64.6, "latency_ms": 15.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|m|tensorrt|int8|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0412-jetson-orin-nx-yolo12-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 178.7, "latency_ms": 5.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|n|tensorrt|fp16|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0405-jetson-orin-nx-yolo12-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 218, "latency_ms": 4.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|n|tensorrt|int8|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0410-jetson-orin-nx-yolo12-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 112.1, "latency_ms": 8.92, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|s|tensorrt|fp16|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0406-jetson-orin-nx-yolo12-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 136.8, "latency_ms": 7.31, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|s|tensorrt|int8|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0411-jetson-orin-nx-yolo12-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 20.5, "latency_ms": 48.78, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|x|tensorrt|fp16|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0409-jetson-orin-nx-yolo12-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "15w", "software": null, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 24.6, "latency_ms": 40.65, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": 15, "source_value": "15w" }, "key": "jetson_orin_nx|yolo12|x|tensorrt|int8|640x640|1|15w", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0414-jetson-orin-nx-yolo12-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 32, "latency_ms": 31.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0097-jetson-orin-nx-yolov8-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 56, "latency_ms": 17.86, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0096-jetson-orin-nx-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 193, "latency_ms": 5.19, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0093-jetson-orin-nx-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 264, "latency_ms": 3.79, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0089-jetson-orin-nx-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 133, "latency_ms": 7.52, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0095-jetson-orin-nx-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx", "power_mode": "maxn", "software": { "name": "jetpack", "version": "YOLOv8 v8.x, JetPack 6.1, TensorRT 8.6.x" }, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 20, "latency_ms": 50, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics Inc.", "document": "https://docs.ultralytics.com/guides/nvidia-jetson/#benchmark-results", "doc_id": "bench:ultralytics-inc-https-docs-ultralytics-com-guides-nvidia-jetson-benchmark-results", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx|yolov8|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0098-jetson-orin-nx-yolov8-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-pose-orin-nx-estimated-fp16-095", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8", "variant": "n-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 175, "latency_ms": 5.71, "latency_p99_ms": 7, "tokens_per_s": null, "power_w": null, "memory_mb": 950, "accuracy_map50_95": 0.504 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-pose-orin-nx-estimated-fp16-095", "notes": "T4 to Orin NX GPU scaling factor ~3.4x for this model class. Pose overhead ~20% vs detection.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-pose-orin-nx-estimated-fp16-095" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|yolov8_pose|n-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0532-jetson-orin-nx-yolov8-pose-n-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-seg-orin-nx-estimated-fp16-093", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8", "variant": "n-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 160, "latency_ms": 6.25, "latency_p99_ms": 7.6, "tokens_per_s": null, "power_w": null, "memory_mb": 1100, "accuracy_map50_95": 0.367 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-seg-orin-nx-estimated-fp16-093", "notes": "Segmentation adds mask head overhead. Interpolated from detection benchmark with ~25% latency penalty.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-seg-orin-nx-estimated-fp16-093" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|yolov8_seg|n-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0530-jetson-orin-nx-yolov8-seg-n-seg-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov9c-orin-nx-estimated-fp16-087", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "yolov9", "series": "yolov9", "variant": "c", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 48, "latency_ms": 20.8, "latency_p99_ms": 25, "tokens_per_s": null, "power_w": null, "memory_mb": 7500, "accuracy_map50_95": 0.53 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov9c-orin-nx-estimated-fp16-087", "notes": "YOLOv9c is ~51% heavier than YOLO11m by FLOPs. mAP50-95=0.530 from YOLOv9 paper.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov9c-orin-nx-estimated-fp16-087" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|yolov9|c|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0524-jetson-orin-nx-yolov9-c-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov9s-orin-nx-estimated-fp16-086", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "yolov9", "series": "yolov9", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 115, "latency_ms": 8.7, "latency_p99_ms": 10.5, "tokens_per_s": null, "power_w": null, "memory_mb": 2200, "accuracy_map50_95": 0.468 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov9s-orin-nx-estimated-fp16-086", "notes": "YOLOv9s ~24% heavier than YOLO11s by FLOPs. mAP50-95=0.468 from YOLOv9 paper.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov9s-orin-nx-estimated-fp16-086" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|yolov9|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0523-jetson-orin-nx-yolov9-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov9t-orin-nx-estimated-fp16-085", "hardware": "jetson_orin_nx", "power_mode": null, "software": null, "model": { "family": "yolov9", "series": "yolov9", "variant": "t", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 195, "latency_ms": 5.13, "latency_p99_ms": 6.2, "tokens_per_s": null, "power_w": null, "memory_mb": 900, "accuracy_map50_95": 0.383 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov9t-orin-nx-estimated-fp16-085", "notes": "YOLOv9t is ~18% heavier than YOLO11n by FLOPs. mAP50-95=0.383 from YOLOv9 paper.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov9t-orin-nx-estimated-fp16-085" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_orin_nx|yolov9|t|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0522-jetson-orin-nx-yolov9-t-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 31.7, "latency_ms": 31.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolo11|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0072-jetson-orin-nx-8gb-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 53.1, "latency_ms": 18.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolo11|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0071-jetson-orin-nx-8gb-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 194.5, "latency_ms": 5.14, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolo11|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0068-jetson-orin-nx-8gb-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 231.6, "latency_ms": 4.32, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolo11|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0069-jetson-orin-nx-8gb-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 127.8, "latency_ms": 7.83, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolo11|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0070-jetson-orin-nx-8gb-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 35, "latency_ms": 28.57, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0172-jetson-orin-nx-8gb-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 50, "latency_ms": 20, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0173-jetson-orin-nx-8gb-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 120, "latency_ms": 8.33, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0168-jetson-orin-nx-8gb-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 160, "latency_ms": 6.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0169-jetson-orin-nx-8gb-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 83, "latency_ms": 12.05, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0170-jetson-orin-nx-8gb-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_orin_nx_8gb", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 113, "latency_ms": 8.85, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_orin_nx_8gb|yolov8|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0171-jetson-orin-nx-8gb-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 3640, "latency_ms": 0.27, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin MobileNetV2 FP16 ~2800 fps × 1.3 (heavily bandwidth-limited). MobileNetV2 is deeply bandwidth-bound — depthwise convolutions. Thor bandwidth advantage limited to ~1.3x. 2800 × 1.3 = 3640.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|mobilenetv2|v2|tensorrt|fp16|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0635-jetson-thor-t5000-mobilenetv2-v2-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 4004, "latency_ms": 0.25, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor MobileNetV2 FP16 3640 fps × 1.1 (bandwidth-limited INT8 gain). MobileNetV2 INT8 gain is minimal at this throughput — bottleneck is memory bandwidth not compute. 3640 × 1.1 = 4004.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|mobilenetv2|v2|tensorrt|int8|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0636-jetson-thor-t5000-mobilenetv2-v2-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2100, "latency_ms": 0.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin ResNet50 FP16 ~1400 fps × 1.5 (bandwidth-limited scaling). ResNet50 is partially memory-bandwidth-bound at 224x224. Thor bandwidth advantage ~1.5x vs Orin. 1400 × 1.5 = 2100.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|resnet|50|tensorrt|fp16|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0632-jetson-thor-t5000-resnet-50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2940, "latency_ms": 0.34, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor ResNet50 FP16 2100 fps × 1.4 (bandwidth-limited FP8 gain). Classification models are bandwidth-limited — FP8 gain reduced to 1.4x. 2100 × 1.4 = 2940.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|resnet|50|tensorrt|fp8|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0634-jetson-thor-t5000-resnet-50-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2520, "latency_ms": 0.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin ResNet50 INT8 ~2100 fps × 1.2 (conservative bandwidth scaling). Classification INT8 benefits more from quantization (larger tensors). 2100 × 1.2 = 2520.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|resnet|50|tensorrt|int8|224x224|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0633-jetson-thor-t5000-resnet-50-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 227, "latency_ms": 4.41, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11l FP16 124 fps × 1.835 Thor/Orin ratio. 124 × 1.835 = 227.5 ≈ 227. YOLO12l FP16 = 299 FPS. Gap: 24%.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0620-jetson-thor-t5000-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 234, "latency_ms": 4.27, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11l FP16 227 fps × 1.031 Blackwell INT8/FP16 ratio. 227 × 1.031 = 234.0.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0621-jetson-thor-t5000-yolo11-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 294, "latency_ms": 3.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11m FP16 160 fps × 1.835 Thor/Orin ratio. 160 × 1.835 = 293.6 ≈ 294. YOLO12m FP16 = 388 FPS. Gap: 24%. Consistent with larger attention acceleration advantage at m-tier.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0618-jetson-thor-t5000-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 470, "latency_ms": 2.13, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11m FP16 294 fps × 1.6 Blackwell FP8 gain. 294 × 1.6 = 470.4 ≈ 470.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|m|tensorrt|fp8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0638-jetson-thor-t5000-yolo11-m-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 305, "latency_ms": 3.28, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11m FP16 294 fps × 1.036 Blackwell INT8/FP16 ratio. 294 × 1.036 = 304.6 ≈ 305.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0619-jetson-thor-t5000-yolo11-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 719, "latency_ms": 1.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11n FP16 392 fps × 1.835 Thor/Orin ratio. AGX Orin YOLO11n FP16 = 392 FPS × 1.835 (Thor/Orin ratio from YOLO26 cross-platform) = 719 FPS. Cross-validated: YOLO12n FP16 = 719 FPS (same GFLOPs 6.5B).", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0614-jetson-thor-t5000-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1150, "latency_ms": 0.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11n FP16 719 fps × 1.6 Blackwell FP8 gain (conservative). FP8 E4M3 on Blackwell: theoretical 2.0x, practical 1.5-1.8x for edge inference. Conservative 1.6x applied. 719 × 1.6 = 1150. TensorRT 10.x FP8 auto-calibration required.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|n|tensorrt|fp8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0637-jetson-thor-t5000-yolo11-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 657, "latency_ms": 1.52, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11n FP16 719 fps × 0.914 Blackwell INT8/FP16 ratio. 719 × 0.914 = 657.2 ≈ 657. Blackwell INT8 slower than FP16 for n-tier. Cross-validated with YOLO26n INT8 = 657.9 FPS.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0615-jetson-thor-t5000-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 503, "latency_ms": 1.99, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11s FP16 274 fps × 1.835 Thor/Orin ratio. 274 × 1.835 = 502.8 ≈ 503. Cross-check: YOLO12s FP16 = 565 FPS (21.4B GFLOPs vs YOLO11s 21.5B). YOLO11s has 0.5% more GFLOPs → expect 0.5% lower FPS. 565 × 0.995 = 562 vs 503 — gap of 11%. YOLO12 attention layers may get Blackwell optimization advantage not available to CNN YOLO11.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0616-jetson-thor-t5000-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 508, "latency_ms": 1.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11s FP16 503 fps × 1.011 Blackwell INT8/FP16 ratio. 503 × 1.011 = 508.5 ≈ 508. INT8 ≈ FP16 at s-tier on Blackwell.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0617-jetson-thor-t5000-yolo11-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 136, "latency_ms": 7.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11x FP16 74 fps × 1.835 Thor/Orin ratio. 74 × 1.835 = 135.8 ≈ 136. YOLO12x FP16 = 193 FPS. Gap: 30%. Attention acceleration advantage grows with model size.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|x|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0622-jetson-thor-t5000-yolo11-x-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 218, "latency_ms": 4.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11x FP16 136 fps × 1.6 Blackwell FP8 gain. 136 × 1.6 = 217.6 ≈ 218.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|x|tensorrt|fp8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0639-jetson-thor-t5000-yolo11-x-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 149, "latency_ms": 6.71, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11x FP16 136 fps × 1.093 Blackwell INT8/FP16 ratio. 136 × 1.093 = 148.6 ≈ 149.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolo11|x|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0623-jetson-thor-t5000-yolo11-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 218, "latency_ms": 4.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11l det 124 fps → pose GFLOPs ratio 90.3/86.9 → Thor 1.835x. Orin YOLO11l det FP16 124 fps / 1.039 pose/det GFLOPs ratio = 119 fps × 1.835 Thor/Orin = 218 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0656-jetson-thor-t5000-yolo11-pose-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 283, "latency_ms": 3.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 283 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|l|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0664-jetson-thor-t5000-yolo11-pose-l-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 254, "latency_ms": 3.94, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 254 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0660-jetson-thor-t5000-yolo11-pose-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 269, "latency_ms": 3.72, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835; the research notes do not record the exact factors behind this 269 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0655-jetson-thor-t5000-yolo11-pose-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 369, "latency_ms": 2.71, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 369 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|m|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0663-jetson-thor-t5000-yolo11-pose-m-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 328, "latency_ms": 3.05, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 328 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0659-jetson-thor-t5000-yolo11-pose-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 629, "latency_ms": 1.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835; the research notes do not record the exact factors behind this 629 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0653-jetson-thor-t5000-yolo11-pose-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 689, "latency_ms": 1.45, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 689 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|n|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0661-jetson-thor-t5000-yolo11-pose-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 575, "latency_ms": 1.74, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 575 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0657-jetson-thor-t5000-yolo11-pose-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 466, "latency_ms": 2.15, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835; the research notes do not record the exact factors behind this 466 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0654-jetson-thor-t5000-yolo11-pose-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 536, "latency_ms": 1.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 536 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|s|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0662-jetson-thor-t5000-yolo11-pose-s-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 498, "latency_ms": 2.01, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 pose on the Thor T5000 scaled from Orin YOLO11 detection throughput by the pose/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 498 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_pose|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0658-jetson-thor-t5000-yolo11-pose-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 150, "latency_ms": 6.67, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11l det 124 fps → seg GFLOPs ratio 132.2/86.9 → Thor 1.835x. Orin YOLO11l det FP16 124 fps / 1.521 seg/det GFLOPs ratio = 82 fps × 1.835 Thor/Orin = 150 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0644-jetson-thor-t5000-yolo11-seg-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 218, "latency_ms": 4.59, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 218 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|l|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0652-jetson-thor-t5000-yolo11-seg-l-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 195, "latency_ms": 5.13, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 195 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0648-jetson-thor-t5000-yolo11-seg-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 217, "latency_ms": 4.61, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835; the research notes do not record the exact factors behind this 217 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0643-jetson-thor-t5000-yolo11-seg-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 295, "latency_ms": 3.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 295 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|m|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0651-jetson-thor-t5000-yolo11-seg-m-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 263, "latency_ms": 3.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 263 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0647-jetson-thor-t5000-yolo11-seg-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 483, "latency_ms": 2.07, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Orin YOLO11n det 392 fps → seg GFLOPs ratio 9.7/6.5 → Thor 1.835x. Orin YOLO11n det FP16 392 fps / 1.492 seg/det GFLOPs ratio = 263 fps × 1.835 Thor/Orin = 483 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0641-jetson-thor-t5000-yolo11-seg-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 530, "latency_ms": 1.89, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 530 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|n|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0649-jetson-thor-t5000-yolo11-seg-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 441, "latency_ms": 2.27, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor seg-n FP16 483 fps × 0.914 Blackwell INT8/FP16. Thor seg-n FP16 483 fps × 0.914 Blackwell INT8/FP16 ratio = 441 fps. INT8 is 8.7% SLOWER than FP16 — inversion holds for seg.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0645-jetson-thor-t5000-yolo11-seg-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 367, "latency_ms": 2.72, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835; the research notes do not record the exact factors behind this 367 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0642-jetson-thor-t5000-yolo11-seg-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 421, "latency_ms": 2.38, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell FP8/FP16 factor; the research notes do not record the exact factors behind this 421 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|s|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0650-jetson-thor-t5000-yolo11-seg-s-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 390, "latency_ms": 2.56, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: YOLO11 segmentation on the Thor T5000 scaled from Orin YOLO11 detection throughput by the segmentation/detection GFLOPs ratio and a Thor/Orin factor of 1.835, then a Blackwell INT8/FP16 factor; the research notes do not record the exact factors behind this 390 fps value.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo11_seg|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0646-jetson-thor-t5000-yolo11-seg-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": { "name": "ultralytics", "version": "8.4.7" }, "model": { "family": "yolo12", "series": "yolo12", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 299, "latency_ms": 3.34, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo12|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0004-jetson-thor-t5000-yolo12-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": { "name": "ultralytics", "version": "8.4.7" }, "model": { "family": "yolo12", "series": "yolo12", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 388, "latency_ms": 2.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo12|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0003-jetson-thor-t5000-yolo12-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": { "name": "ultralytics", "version": "8.4.7" }, "model": { "family": "yolo12", "series": "yolo12", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 719, "latency_ms": 1.39, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo12|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0001-jetson-thor-t5000-yolo12-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": { "name": "ultralytics", "version": "8.4.7" }, "model": { "family": "yolo12", "series": "yolo12", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 565, "latency_ms": 1.77, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo12|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0002-jetson-thor-t5000-yolo12-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": { "name": "ultralytics", "version": "8.4.7" }, "model": { "family": "yolo12", "series": "yolo12", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 193, "latency_ms": 5.18, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": null, "sources": [ { "publisher": "Ultralytics", "document": "Publisher benchmark table", "doc_id": "bench:ultralytics-publisher-benchmark-table", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolo12|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0005-jetson-thor-t5000-yolo12-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26l-agx-thor-fp16-122", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 299.4, "latency_ms": 3.34, "latency_p99_ms": 4, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.6243 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT FP16.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0549-jetson-thor-t5000-yolo26-l-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26l-agx-thor-int8-123", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 308.6, "latency_ms": 3.24, "latency_p99_ms": 3.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.5708 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT INT8.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0550-jetson-thor-t5000-yolo26-l-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolo26m-agx-thor-fp16-120", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 387.6, "latency_ms": 2.58, "latency_p99_ms": 3.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.6209 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT FP16.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0547-jetson-thor-t5000-yolo26-m-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26m-agx-thor-int8-121", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 401.6, "latency_ms": 2.49, "latency_p99_ms": 3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.5595 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT INT8.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0548-jetson-thor-t5000-yolo26-m-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolo26n-agx-thor-fp16-116", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 719.4, "latency_ms": 1.39, "latency_p99_ms": 1.7, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.4797 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT FP16, COCO val2017. AGX Thor T5000 with Blackwell GPU.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0543-jetson-thor-t5000-yolo26-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26n-agx-thor-int8-117", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 657.9, "latency_ms": 1.52, "latency_p99_ms": 1.85, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.4273 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT INT8, COCO val2017.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0544-jetson-thor-t5000-yolo26-n-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolo26s-agx-thor-fp16-118", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 564.9, "latency_ms": 1.77, "latency_p99_ms": 2.15, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.565 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT FP16, COCO val2017.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0545-jetson-thor-t5000-yolo26-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26s-agx-thor-int8-119", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 571.4, "latency_ms": 1.75, "latency_p99_ms": 2.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.501 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT INT8, COCO val2017.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0546-jetson-thor-t5000-yolo26-s-tensorrt-int8" }, { "origin": "research_corpus", "record_id": "yolo26x-agx-thor-fp16-124", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 193.8, "latency_ms": 5.16, "latency_p99_ms": 6.2, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.6611 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT FP16.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0551-jetson-thor-t5000-yolo26-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo26x-agx-thor-int8-125", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 211.9, "latency_ms": 4.72, "latency_p99_ms": 5.7, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.5222 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "TensorRT INT8.", "sources": [ { "publisher": "Ultralytics", "document": "NVIDIA Jetson guide — YOLO benchmarks", "doc_id": "ultralytics:guides/nvidia-jetson", "url": "https://docs.ultralytics.com/guides/nvidia-jetson/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "jetson_thor_t5000|yolo26|x|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0552-jetson-thor-t5000-yolo26-x-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 119, "latency_ms": 8.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11l FP16 227 fps × (86.9/165.2) GFLOPs ratio. 227 × (86.9/165.2) = 119.4 ≈ 119.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|l|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0630-jetson-thor-t5000-yolov8-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 123, "latency_ms": 8.13, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l FP16 119 fps × 1.031 Blackwell INT8/FP16 ratio. 119 × 1.031 = 122.7 ≈ 123.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|l|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0631-jetson-thor-t5000-yolov8-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 253, "latency_ms": 3.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11m FP16 294 fps × (68.0/78.9) GFLOPs ratio. 294 × (68.0/78.9) = 253.4 ≈ 253.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|m|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0628-jetson-thor-t5000-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 262, "latency_ms": 3.82, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m FP16 253 fps × 1.036 Blackwell INT8/FP16 ratio. 253 × 1.036 = 262.1 ≈ 262.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|m|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0629-jetson-thor-t5000-yolov8-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 537, "latency_ms": 1.86, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11n FP16 719 fps × (6.5/8.7) GFLOPs ratio. YOLO11n FP16 = 719 FPS. YOLOv8n has 8.7B GFLOPs vs YOLO11n 6.5B. 719 × (6.5/8.7) = 537.2 ≈ 537.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|n|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0624-jetson-thor-t5000-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 859, "latency_ms": 1.16, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n FP16 537 fps × 1.6 Blackwell FP8 gain (conservative). 537 × 1.6 = 859.2 ≈ 859.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|n|tensorrt|fp8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0640-jetson-thor-t5000-yolov8-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 491, "latency_ms": 2.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n FP16 537 fps × 0.914 Blackwell INT8/FP16 ratio. 537 × 0.914 = 490.8 ≈ 491.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|n|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0625-jetson-thor-t5000-yolov8-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 378, "latency_ms": 2.65, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLO11s FP16 503 fps × (21.5/28.6) GFLOPs ratio. 503 × (21.5/28.6) = 378.0.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|s|tensorrt|fp16|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0626-jetson-thor-t5000-yolov8-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": "maxn", "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 382, "latency_ms": 2.62, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s FP16 378 fps × 1.011 Blackwell INT8/FP16 ratio. 378 × 1.011 = 382.2 ≈ 382.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "preset", "watts": null, "source_value": "maxn" }, "key": "jetson_thor_t5000|yolov8|s|tensorrt|int8|640x640|1|maxn", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0627-jetson-thor-t5000-yolov8-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 117, "latency_ms": 8.55, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l det 119 fps -> pose GFLOPs ratio 168.6/165.2. Thor YOLOv8l det FP16 119 fps / 1.021 pose/det GFLOPs ratio (168.6/165.2) = 117 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0681-jetson-thor-t5000-yolov8-pose-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 187, "latency_ms": 5.35, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l-pose FP16 117 fps x 1.6 FP8 gain. Thor pose-l FP16 117 fps x 1.6 Blackwell FP8 gain = 187 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|l|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0689-jetson-thor-t5000-yolov8-pose-l-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "l", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 121, "latency_ms": 8.26, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l-pose FP16 117 fps x 1.031 Blackwell INT8/FP16. Thor pose-l FP16 117 fps x 1.031 Blackwell INT8/FP16 ratio = 121 fps. INT8 3.4% faster than FP16.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0685-jetson-thor-t5000-yolov8-pose-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 246, "latency_ms": 4.07, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m det 253 fps -> pose GFLOPs ratio 81.0/78.9. Thor YOLOv8m det FP16 253 fps / 1.027 pose/det GFLOPs ratio (81.0/78.9) = 246 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0680-jetson-thor-t5000-yolov8-pose-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 394, "latency_ms": 2.54, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m-pose FP16 246 fps x 1.6 FP8 gain. Thor pose-m FP16 246 fps x 1.6 Blackwell FP8 gain = 394 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|m|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0688-jetson-thor-t5000-yolov8-pose-m-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "m", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 255, "latency_ms": 3.92, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m-pose FP16 246 fps x 1.036 Blackwell INT8/FP16. Thor pose-m FP16 246 fps x 1.036 Blackwell INT8/FP16 ratio = 255 fps. INT8 3.7% faster than FP16.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0684-jetson-thor-t5000-yolov8-pose-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 508, "latency_ms": 1.97, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n det 537 fps -> pose GFLOPs ratio 9.2/8.7. Thor YOLOv8n det FP16 537 fps / 1.057 pose/det GFLOPs ratio (9.2/8.7) = 508 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0678-jetson-thor-t5000-yolov8-pose-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 813, "latency_ms": 1.23, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n-pose FP16 508 fps x 1.6 FP8 gain. Thor pose-n FP16 508 fps x 1.6 Blackwell FP8 gain = 813 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|n|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0686-jetson-thor-t5000-yolov8-pose-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 464, "latency_ms": 2.16, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n-pose FP16 508 fps x 0.914 Blackwell INT8/FP16. Thor pose-n FP16 508 fps x 0.914 Blackwell INT8/FP16 ratio = 464 fps. INT8 is 8.7% SLOWER than FP16 — inversion holds for YOLOv8 pose.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0682-jetson-thor-t5000-yolov8-pose-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 358, "latency_ms": 2.79, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s det 378 fps -> pose GFLOPs ratio 30.2/28.6. Thor YOLOv8s det FP16 378 fps / 1.056 pose/det GFLOPs ratio (30.2/28.6) = 358 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0679-jetson-thor-t5000-yolov8-pose-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 573, "latency_ms": 1.75, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s-pose FP16 358 fps x 1.6 FP8 gain. Thor pose-s FP16 358 fps x 1.6 Blackwell FP8 gain = 573 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|s|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0687-jetson-thor-t5000-yolov8-pose-s-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 362, "latency_ms": 2.76, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s-pose FP16 358 fps x 1.011 Blackwell INT8/FP16. Thor pose-s FP16 358 fps x 1.011 Blackwell INT8/FP16 ratio = 362 fps. INT8 ~= FP16 at s-tier.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_pose|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0683-jetson-thor-t5000-yolov8-pose-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 89, "latency_ms": 11.24, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l det 119 fps -> seg GFLOPs ratio 220.5/165.2. Thor YOLOv8l det FP16 119 fps / 1.335 seg/det GFLOPs ratio (220.5/165.2) = 89 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0669-jetson-thor-t5000-yolov8-seg-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 142, "latency_ms": 7.04, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l-seg FP16 89 fps x 1.6 FP8 gain. Thor seg-l FP16 89 fps x 1.6 Blackwell FP8 gain = 142 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|l|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0677-jetson-thor-t5000-yolov8-seg-l-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "l", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 92, "latency_ms": 10.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8l-seg FP16 89 fps x 1.031 Blackwell INT8/FP16. Thor seg-l FP16 89 fps x 1.031 Blackwell INT8/FP16 ratio = 92 fps. INT8 3.4% faster than FP16.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|l|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0673-jetson-thor-t5000-yolov8-seg-l-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 181, "latency_ms": 5.52, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m det 253 fps -> seg GFLOPs ratio 110.2/78.9. Thor YOLOv8m det FP16 253 fps / 1.397 seg/det GFLOPs ratio (110.2/78.9) = 181 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0668-jetson-thor-t5000-yolov8-seg-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 290, "latency_ms": 3.45, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m-seg FP16 181 fps x 1.6 FP8 gain. Thor seg-m FP16 181 fps x 1.6 Blackwell FP8 gain = 290 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|m|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0676-jetson-thor-t5000-yolov8-seg-m-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 188, "latency_ms": 5.32, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8m-seg FP16 181 fps x 1.036 Blackwell INT8/FP16. Thor seg-m FP16 181 fps x 1.036 Blackwell INT8/FP16 ratio = 188 fps. INT8 3.9% faster than FP16.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|m|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0672-jetson-thor-t5000-yolov8-seg-m-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 371, "latency_ms": 2.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n det 537 fps -> seg GFLOPs ratio 12.6/8.7. Thor YOLOv8n det FP16 537 fps / 1.448 seg/det GFLOPs ratio (12.6/8.7) = 371 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0666-jetson-thor-t5000-yolov8-seg-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 594, "latency_ms": 1.68, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n-seg FP16 371 fps x 1.6 FP8 gain. Thor seg-n FP16 371 fps x 1.6 Blackwell FP8 gain = 594 fps. Seg mask upsampling is partially bandwidth-bound — conservative estimate.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|n|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0674-jetson-thor-t5000-yolov8-seg-n-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 339, "latency_ms": 2.95, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8n-seg FP16 371 fps x 0.914 Blackwell INT8/FP16. Thor seg-n FP16 371 fps x 0.914 Blackwell INT8/FP16 ratio = 339 fps. INT8 is 8.6% SLOWER than FP16 — inversion holds for YOLOv8 seg.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0670-jetson-thor-t5000-yolov8-seg-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 254, "latency_ms": 3.94, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s det 378 fps -> seg GFLOPs ratio 42.6/28.6. Thor YOLOv8s det FP16 378 fps / 1.490 seg/det GFLOPs ratio (42.6/28.6) = 254 fps", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0667-jetson-thor-t5000-yolov8-seg-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 406, "latency_ms": 2.46, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s-seg FP16 254 fps x 1.6 FP8 gain. Thor seg-s FP16 254 fps x 1.6 Blackwell FP8 gain = 406 fps.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|s|tensorrt|fp8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0675-jetson-thor-t5000-yolov8-seg-s-tensorrt-fp8" }, { "origin": "catalog", "hardware": "jetson_thor_t5000", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 257, "latency_ms": 3.89, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured: Thor YOLOv8s-seg FP16 254 fps x 1.011 Blackwell INT8/FP16. Thor seg-s FP16 254 fps x 1.011 Blackwell INT8/FP16 ratio = 257 fps. INT8 ~= FP16 at s-tier.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "measured_w", "watts": 130, "source_value": "130w" }, "key": "jetson_thor_t5000|yolov8_seg|s|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0671-jetson-thor-t5000-yolov8-seg-s-tensorrt-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 3, "latency_ms": 333.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|l|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0395-neousys-nuvo9531-yolo11-l-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "fp16", "accelerator": "igpu", "scope": "inference_only", "metrics": { "fps": 6, "latency_ms": 166.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|m|openvino|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0404-neousys-nuvo9531-yolo11-m-openvino-fp16" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 4, "latency_ms": 250, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|m|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0394-neousys-nuvo9531-yolo11-m-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 20, "latency_ms": 50, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|n|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0333-neousys-nuvo9531-yolo11-n-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 9, "latency_ms": 111.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|s|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0393-neousys-nuvo9531-yolo11-s-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 1.8, "latency_ms": 555.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11|x|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0396-neousys-nuvo9531-yolo11-x-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 3, "latency_ms": 333.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11_seg|m|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0402-neousys-nuvo9531-yolo11-seg-m-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 15, "latency_ms": 66.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11_seg|n|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0400-neousys-nuvo9531-yolo11-seg-n-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 6.5, "latency_ms": 153.8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolo11_seg|s|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0401-neousys-nuvo9531-yolo11-seg-s-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 2.5, "latency_ms": 400, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|l|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0391-neousys-nuvo9531-yolov5-l-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 5, "latency_ms": 200, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|m|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0390-neousys-nuvo9531-yolov5-m-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "fp16", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 30, "latency_ms": 33.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "low", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|n|openvino|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0336-neousys-nuvo9531-yolov5-n-openvino-fp16" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 22, "latency_ms": 45.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|n|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0334-neousys-nuvo9531-yolov5-n-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 12, "latency_ms": 83.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|s|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0389-neousys-nuvo9531-yolov5-s-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 1.5, "latency_ms": 666.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov5|x|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0392-neousys-nuvo9531-yolov5-x-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 2.5, "latency_ms": 400, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|l|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0387-neousys-nuvo9531-yolov8-l-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "fp16", "accelerator": "igpu", "scope": "inference_only", "metrics": { "fps": 6, "latency_ms": 166.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|m|openvino|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0403-neousys-nuvo9531-yolov8-m-openvino-fp16" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 4.5, "latency_ms": 222.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|m|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0386-neousys-nuvo9531-yolov8-m-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "fp16", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 28, "latency_ms": 35.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "low", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|n|openvino|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0335-neousys-nuvo9531-yolov8-n-openvino-fp16" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 18, "latency_ms": 55.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|n|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0331-neousys-nuvo9531-yolov8-n-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 10, "latency_ms": 100, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Publisher", "document": "Ultralytics OpenVINO Integration — YOLO11 Benchmarks on Intel CPUs", "doc_id": "bench:publisher-ultralytics-openvino-integration-yolo11-benchmarks-on-intel-cpus", "url": "https://docs.ultralytics.com/integrations/openvino/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|s|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0332-neousys-nuvo9531-yolov8-s-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 2, "latency_ms": 500, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8|x|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0388-neousys-nuvo9531-yolov8-x-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 3, "latency_ms": 333.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8_seg|m|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0399-neousys-nuvo9531-yolov8-seg-m-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 14, "latency_ms": 71.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8_seg|n|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0397-neousys-nuvo9531-yolov8-seg-n-openvino-int8" }, { "origin": "catalog", "hardware": "neousys_nuvo9531", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "openvino", "precision": "int8", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 6, "latency_ms": 166.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": 15, "source_value": "15w" }, "key": "neousys_nuvo9531|yolov8_seg|s|openvino|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0398-neousys-nuvo9531-yolov8-seg-s-openvino-int8" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 7200, "latency_ms": 0.14, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|resnet|50|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0444-nvidia-a10-server-resnet-50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 173, "latency_ms": 5.78, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0441-nvidia-a10-server-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 224, "latency_ms": 4.46, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0440-nvidia-a10-server-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 627, "latency_ms": 1.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0438-nvidia-a10-server-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 438, "latency_ms": 2.28, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolo11|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0439-nvidia-a10-server-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 214, "latency_ms": 4.67, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0443-nvidia-a10-server-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a10_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 594, "latency_ms": 1.68, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a10_server|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0442-nvidia-a10-server-yolov8-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8l-a100-fp16-035", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 418.4, "latency_ms": 2.39, "latency_p99_ms": 3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.529 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0478-nvidia-a100-yolov8-l-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8m-a100-fp16-034", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 546.4, "latency_ms": 1.83, "latency_p99_ms": 2.3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.502 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0477-nvidia-a100-yolov8-m-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-a100-fp16-032", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1010.1, "latency_ms": 0.99, "latency_p99_ms": 1.3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.373 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017, reference baseline for scaling", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0475-nvidia-a100-yolov8-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8s-a100-fp16-033", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 833.3, "latency_ms": 1.2, "latency_p99_ms": 1.6, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.449 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0476-nvidia-a100-yolov8-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8x-a100-fp16-036", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 283.3, "latency_ms": 3.53, "latency_p99_ms": 4.4, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.539 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0479-nvidia-a100-yolov8-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-pose-a100-fp16-039", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8", "variant": "n-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 847.5, "latency_ms": 1.18, "latency_p99_ms": 1.5, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.504 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO pose val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8_pose|n-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0482-nvidia-a100-yolov8-pose-n-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8s-pose-a100-fp16-040", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8", "variant": "s-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 704.2, "latency_ms": 1.42, "latency_p99_ms": 1.8, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.6 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO pose val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8_pose|s-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0483-nvidia-a100-yolov8-pose-s-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8n-seg-a100-fp16-037", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8", "variant": "n-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 826.4, "latency_ms": 1.21, "latency_p99_ms": 1.6, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.367 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017, mAP is box mAP", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8_seg|n-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0480-nvidia-a100-yolov8-seg-n-seg-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov8s-seg-a100-fp16-038", "hardware": "nvidia_a100", "power_mode": null, "software": null, "model": { "family": "yolov8_seg", "series": "yolov8", "variant": "s-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 680.3, "latency_ms": 1.47, "latency_p99_ms": 1.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.446 }, "class": "C", "confidence": "high", "date": "2023-Q3", "derivation": null, "notes": "TensorRT on A100, COCO val2017, mAP is box mAP", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv8 model page — performance table", "doc_id": "ultralytics:models/yolov8", "url": "https://docs.ultralytics.com/models/yolov8/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a100|yolov8_seg|s-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0481-nvidia-a100-yolov8-seg-s-seg-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a2_server", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 398, "latency_ms": 2.51, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a2_server|efficientdet|d0|tensorrt|fp16|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0437-nvidia-a2-server-efficientdet-d0-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a2_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 152, "latency_ms": 6.58, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a2_server|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0434-nvidia-a2-server-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a2_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 373, "latency_ms": 2.68, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a2_server|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0433-nvidia-a2-server-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a2_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 143, "latency_ms": 6.99, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a2_server|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0436-nvidia-a2-server-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_a2_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 352, "latency_ms": 2.84, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_a2_server|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0435-nvidia-a2-server-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 920, "latency_ms": 1.09, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|efficientdet|d0|tensorrt|fp16|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0427-nvidia-l4-server-efficientdet-d0-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 4800, "latency_ms": 0.21, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|resnet|50|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0428-nvidia-l4-server-resnet-50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 223, "latency_ms": 4.48, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0424-nvidia-l4-server-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 288, "latency_ms": 3.47, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0423-nvidia-l4-server-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 862, "latency_ms": 1.16, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0420-nvidia-l4-server-yolo11-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "int8", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 1150, "latency_ms": 0.87, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolo11|n|tensorrt|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0421-nvidia-l4-server-yolo11-n-tensorrt-int8" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 602, "latency_ms": 1.66, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolo11|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0422-nvidia-l4-server-yolo11-s-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 272, "latency_ms": 3.68, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolov8|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0426-nvidia-l4-server-yolov8-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l4_server", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 816, "latency_ms": 1.23, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l4_server|yolov8|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0425-nvidia-l4-server-yolov8-n-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l40s_server", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 15360, "latency_ms": 0.065, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l40s_server|resnet|50|tensorrt|fp16|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0432-nvidia-l40s-server-resnet-50-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l40s_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 714, "latency_ms": 1.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l40s_server|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0431-nvidia-l40s-server-yolo11-l-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l40s_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 921, "latency_ms": 1.09, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l40s_server|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0430-nvidia-l40s-server-yolo11-m-tensorrt-fp16" }, { "origin": "catalog", "hardware": "nvidia_l40s_server", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 2760, "latency_ms": 0.36, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "high", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_l40s_server|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0429-nvidia-l40s-server-yolo11-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "rtdetr-l-t4-fp16-041", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "rtdetr", "series": "rt-detr", "variant": "rtdetr-l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 114, "latency_ms": 8.77, "latency_p99_ms": 11, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.53 }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "RT-DETR model page — performance table", "doc_id": "ultralytics:models/rtdetr", "url": "https://docs.ultralytics.com/models/rtdetr/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|rtdetr|rtdetr-l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0484-nvidia-t4-rtdetr-rtdetr-l-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "rtdetr-x-t4-fp16-042", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "rtdetr", "series": "rt-detr", "variant": "rtdetr-x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 74, "latency_ms": 13.51, "latency_p99_ms": 17, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.548 }, "class": "C", "confidence": "high", "date": "2024-Q1", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "RT-DETR model page — performance table", "doc_id": "ultralytics:models/rtdetr", "url": "https://docs.ultralytics.com/models/rtdetr/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|rtdetr|rtdetr-x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0485-nvidia-t4-rtdetr-rtdetr-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11l-t4-fp16-052", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 161.3, "latency_ms": 6.2, "latency_p99_ms": 7.7, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.534 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0495-nvidia-t4-yolo11-l-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11m-t4-fp16-051", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 212.8, "latency_ms": 4.7, "latency_p99_ms": 5.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.515 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0494-nvidia-t4-yolo11-m-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11n-t4-fp16-049", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 666.7, "latency_ms": 1.5, "latency_p99_ms": 1.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.395 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0492-nvidia-t4-yolo11-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11s-t4-fp16-050", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 400, "latency_ms": 2.5, "latency_p99_ms": 3.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.47 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0493-nvidia-t4-yolo11-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11x-t4-fp16-053", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 88.5, "latency_ms": 11.3, "latency_p99_ms": 14.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.547 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0496-nvidia-t4-yolo11-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11m-pose-t4-fp16-057", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11", "variant": "m-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 204.1, "latency_ms": 4.9, "latency_p99_ms": 6.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.649 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO pose val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11_pose|m-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0500-nvidia-t4-yolo11-pose-m-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11n-pose-t4-fp16-056", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11", "variant": "n-pose", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 588.2, "latency_ms": 1.7, "latency_p99_ms": 2.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.5 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO pose val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11_pose|n-pose|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0499-nvidia-t4-yolo11-pose-n-pose-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11m-seg-t4-fp16-055", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11", "variant": "m-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 158.7, "latency_ms": 6.3, "latency_p99_ms": 7.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.515 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017, box mAP50-95", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11_seg|m-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0498-nvidia-t4-yolo11-seg-m-seg-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolo11n-seg-t4-fp16-054", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11", "variant": "n-seg", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 555.6, "latency_ms": 1.8, "latency_p99_ms": 2.3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.389 }, "class": "C", "confidence": "high", "date": "2024-Q4", "derivation": null, "notes": "TensorRT10 FP16 on T4 GPU, COCO val2017, box mAP50-95", "sources": [ { "publisher": "Ultralytics", "document": "YOLO11 model page — performance table", "doc_id": "ultralytics:models/yolo11", "url": "https://docs.ultralytics.com/models/yolo11/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolo11_seg|n-seg|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0497-nvidia-t4-yolo11-seg-n-seg-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10b-t4-fp16-046", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "b", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 174.2, "latency_ms": 5.74, "latency_p99_ms": 7.2, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.525 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|b|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0489-nvidia-t4-yolov10-b-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10l-t4-fp16-047", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 137.4, "latency_ms": 7.28, "latency_p99_ms": 9.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.532 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|l|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0490-nvidia-t4-yolov10-l-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10m-t4-fp16-045", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 211, "latency_ms": 4.74, "latency_p99_ms": 5.9, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.511 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|m|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0488-nvidia-t4-yolov10-m-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10n-t4-fp16-043", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 543.5, "latency_ms": 1.84, "latency_p99_ms": 2.3, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.385 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|n|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0486-nvidia-t4-yolov10-n-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10s-t4-fp16-044", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 401.6, "latency_ms": 2.49, "latency_p99_ms": 3.1, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.463 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|s|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0487-nvidia-t4-yolov10-s-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "yolov10x-t4-fp16-048", "hardware": "nvidia_t4", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp16", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 93.5, "latency_ms": 10.7, "latency_p99_ms": 13.4, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.544 }, "class": "C", "confidence": "high", "date": "2024-Q2", "derivation": null, "notes": "TensorRT FP16 on T4 GPU, COCO val2017", "sources": [ { "publisher": "Ultralytics", "document": "YOLOv10 model page — performance table", "doc_id": "ultralytics:models/yolov10", "url": "https://docs.ultralytics.com/models/yolov10/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_t4|yolov10|x|tensorrt|fp16|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0491-nvidia-t4-yolov10-x-tensorrt-fp16" }, { "origin": "research_corpus", "record_id": "efficientdet-d0-v100-fp32-058", "hardware": "nvidia_v100", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d0", "task": "detection" }, "input": { "resolution": "512x512", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 97, "latency_ms": 10.2, "latency_p99_ms": 13, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.346 }, "class": "C", "confidence": "high", "date": "2020-Q1", "derivation": null, "notes": "Tesla V100, no TensorRT, COCO test-dev, batch=1", "sources": [ { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_v100|efficientdet|d0|tensorrt|fp32|512x512|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0501-nvidia-v100-efficientdet-d0-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "efficientdet-d1-v100-fp32-059", "hardware": "nvidia_v100", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d1", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 74, "latency_ms": 13.5, "latency_p99_ms": 17, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.405 }, "class": "C", "confidence": "high", "date": "2020-Q1", "derivation": null, "notes": "Tesla V100, no TensorRT, COCO test-dev, batch=1", "sources": [ { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_v100|efficientdet|d1|tensorrt|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0502-nvidia-v100-efficientdet-d1-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "efficientdet-d2-v100-fp32-060", "hardware": "nvidia_v100", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d2", "task": "detection" }, "input": { "resolution": "768x768", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 57, "latency_ms": 17.7, "latency_p99_ms": 22, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.43 }, "class": "C", "confidence": "high", "date": "2020-Q1", "derivation": null, "notes": "Tesla V100, no TensorRT, COCO test-dev, batch=1", "sources": [ { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_v100|efficientdet|d2|tensorrt|fp32|768x768|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0503-nvidia-v100-efficientdet-d2-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "efficientdet-d3-v100-fp32-061", "hardware": "nvidia_v100", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d3", "task": "detection" }, "input": { "resolution": "896x896", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 36, "latency_ms": 28, "latency_p99_ms": 35, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.475 }, "class": "C", "confidence": "high", "date": "2020-Q1", "derivation": null, "notes": "Tesla V100, no TensorRT, COCO test-dev, batch=1", "sources": [ { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_v100|efficientdet|d3|tensorrt|fp32|896x896|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0504-nvidia-v100-efficientdet-d3-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "efficientdet-d4-v100-fp32-062", "hardware": "nvidia_v100", "power_mode": null, "software": null, "model": { "family": "efficientdet", "series": "efficientdet", "variant": "d4", "task": "detection" }, "input": { "resolution": "1024x1024", "batch": 1 }, "runtime": "tensorrt", "runtime_inferred": true, "precision": "fp32", "accelerator": "gpu", "scope": "inference_only", "metrics": { "fps": 23, "latency_ms": 42.8, "latency_p99_ms": 53, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": 0.497 }, "class": "C", "confidence": "high", "date": "2020-Q1", "derivation": null, "notes": "Tesla V100, no TensorRT, COCO test-dev, batch=1", "sources": [ { "publisher": "Google AutoML", "document": "EfficientDet repository benchmark table", "doc_id": "github:google/automl/efficientdet", "url": "https://github.com/google/automl/tree/master/efficientdet", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "nvidia_v100|efficientdet|d4|tensorrt|fp32|1024x1024|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0505-nvidia-v100-efficientdet-d4-tensorrt-fp32" }, { "origin": "research_corpus", "record_id": "rpi5-yolo11n-onnx-fp32-152", "hardware": "raspberry_pi_5", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "cpu", "runtime_inferred": true, "precision": "fp32", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 6.8, "latency_ms": 147.2, "latency_p99_ms": 185, "tokens_per_s": null, "power_w": 5, "memory_mb": 512, "accuracy_map50_95": 0.395 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "ONNX format, FP32, Raspberry Pi 5, CPU only. YOLO26n is ~15% faster. mAP50-95=0.395.", "sources": [ { "publisher": "Ultralytics", "document": "Raspberry Pi guide — YOLO benchmarks", "doc_id": "ultralytics:guides/raspberry-pi", "url": "https://docs.ultralytics.com/guides/raspberry-pi/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "raspberry_pi_5|yolo11|n|cpu|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0568-raspberry-pi-5-yolo11-n-cpu-fp32" }, { "origin": "research_corpus", "record_id": "rpi5-yolo26n-onnx-fp32-148", "hardware": "raspberry_pi_5", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "cpu", "runtime_inferred": true, "precision": "fp32", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 7.8, "latency_ms": 128.4, "latency_p99_ms": 160, "tokens_per_s": null, "power_w": 5, "memory_mb": 512, "accuracy_map50_95": 0.4798 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "ONNX format, FP32, Raspberry Pi 5 (8GB ARM Cortex-A76), CPU only. mAP50-95=0.4798.", "sources": [ { "publisher": "Ultralytics", "document": "Raspberry Pi guide — YOLO benchmarks", "doc_id": "ultralytics:guides/raspberry-pi", "url": "https://docs.ultralytics.com/guides/raspberry-pi/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "raspberry_pi_5|yolo26|n|cpu|fp32|640x640|1", "duplicates": [ { "origin": "research_corpus", "class": "C", "fps": 14.8, "tokens_per_s": null, "source": "ultralytics:guides/raspberry-pi" }, { "origin": "research_corpus", "class": "C", "fps": 14.1, "tokens_per_s": null, "source": "ultralytics:guides/raspberry-pi" } ], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0566-raspberry-pi-5-yolo26-n-cpu-fp32" }, { "origin": "research_corpus", "record_id": "rpi5-yolo26s-onnx-fp32-151", "hardware": "raspberry_pi_5", "power_mode": null, "software": null, "model": { "family": "yolo26", "series": "yolo26", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "cpu", "runtime_inferred": true, "precision": "fp32", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 2.8, "latency_ms": 352, "latency_p99_ms": 440, "tokens_per_s": null, "power_w": 6, "memory_mb": 1400, "accuracy_map50_95": 0.5665 }, "class": "C", "confidence": "high", "date": "2026-Q1", "derivation": null, "notes": "ONNX format, FP32, Raspberry Pi 5, CPU only. mAP50-95=0.5665.", "sources": [ { "publisher": "Ultralytics", "document": "Raspberry Pi guide — YOLO benchmarks", "doc_id": "ultralytics:guides/raspberry-pi", "url": "https://docs.ultralytics.com/guides/raspberry-pi/", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "raspberry_pi_5|yolo26|s|cpu|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0567-raspberry-pi-5-yolo26-s-cpu-fp32" }, { "origin": "research_corpus", "record_id": "yolov8n-cpu-onnx-fp32-079", "hardware": "raspberry_pi_5", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "cpu", "runtime_inferred": true, "precision": "fp32", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 12.4, "latency_ms": 80.4, "latency_p99_ms": 98, "tokens_per_s": null, "power_w": 5, "memory_mb": 420, "accuracy_map50_95": 0.373 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8n-cpu-onnx-fp32-079", "notes": "Ultralytics CPU ONNX baseline is ~80ms on modern x86; RPi5 ARM A76 @ 2.4GHz is roughly similar throughput. Estimated from ONNX runtime reference numbers.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8n-cpu-onnx-fp32-079" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "raspberry_pi_5|yolov8|n|cpu|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0520-raspberry-pi-5-yolov8-n-cpu-fp32" }, { "origin": "research_corpus", "record_id": "yolov8s-cpu-onnx-fp32-080", "hardware": "raspberry_pi_5", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "cpu", "runtime_inferred": true, "precision": "fp32", "accelerator": "cpu", "scope": "inference_only", "metrics": { "fps": 5.8, "latency_ms": 172, "latency_p99_ms": 210, "tokens_per_s": null, "power_w": 5.5, "memory_mb": 650, "accuracy_map50_95": 0.449 }, "class": "D", "confidence": "medium", "date": "2024-Q1", "derivation": "EdgeAIStack estimate — corpus record yolov8s-cpu-onnx-fp32-080", "notes": "RPi5 CPU-only, ONNX runtime FP32. Scaled from x86 reference: RPi5 ~1.35x slower than modern x86 laptop for ONNX inference.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "corpus record yolov8s-cpu-onnx-fp32-080" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "raspberry_pi_5|yolov8|s|cpu|fp32|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": "derived", "latency_p99_basis": "modelled_p50_multiplier_band_1.15-1.5", "id": "b0521-raspberry-pi-5-yolov8-s-cpu-fp32" }, { "origin": "catalog", "hardware": "rk3568", "power_mode": null, "software": { "name": "sdk", "version": "rknn_model_zoo v2.3.2" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 180.7, "latency_ms": 5.53, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": "Publisher table \"Model performance benchmark(FPS)\", column \"RK3566 RK3568\" (one figure published for both SoCs); mobilenetv2-12 (ONNX model zoo MobileNet v2-1.0), dtype INT8, inputs_shape [1, 3, 224, 224]. Publisher notes: collected at the maximum NPU frequency of each platform; model inference time only, pre- and post-processing excluded. latency_ms = 1000 / fps.", "sources": [ { "publisher": "Rockchip", "document": "Rockchip / airockchip (official GitHub org) RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "revision": "rknn_model_zoo v2.3.2", "url": "https://github.com/airockchip/rknn_model_zoo/blob/v2.3.2/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3568|mobilenetv2|v2|rknn|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0690-rk3568-mobilenetv2-v2-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "deeplabv3", "series": "deeplabv3", "variant": "mobilenetv2", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 34, "latency_ms": 29.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|deeplabv3|mobilenetv2|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0246-rk3588-npu-deeplabv3-mobilenetv2-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "mobilenet", "series": "mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 350, "latency_ms": 2.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|mobilenet|v1|rknn|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0237-rk3588-npu-mobilenet-v1-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": { "name": "sdk", "version": "rknn_model_zoo v2.3.2" }, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 450.7, "latency_ms": 2.22, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "high", "date": "2025", "derivation": null, "notes": "Publisher table \"Model performance benchmark(FPS)\", column \"RK3588 @single_core\" (one of the three NPU cores); mobilenetv2-12 (ONNX model zoo MobileNet v2-1.0), dtype INT8, inputs_shape [1, 3, 224, 224]. Publisher notes: collected at the maximum NPU frequency of each platform; model inference time only, pre- and post-processing excluded. latency_ms = 1000 / fps.", "sources": [ { "publisher": "Rockchip", "document": "Rockchip / airockchip (official GitHub org) RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "revision": "rknn_model_zoo v2.3.2", "url": "https://github.com/airockchip/rknn_model_zoo/blob/v2.3.2/README.md", "retrieved": "2026-09-28", "verified": "2026-09-28", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|mobilenetv2|v2|rknn|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0236-rk3588-npu-mobilenetv2-v2-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "resnet", "series": "resnet", "variant": "50", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 104.6, "latency_ms": 9.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|resnet|50|rknn|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0235-rk3588-npu-resnet-50-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "retinaface", "series": "retinaface", "variant": "mobile", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 227.2, "latency_ms": 4.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "medium", "date": "2026", "derivation": "EdgeAIStack research ingestion — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|retinaface|mobile|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0245-rk3588-npu-retinaface-mobile-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "ssd_mobilenet", "series": "ssd_mobilenet", "variant": "v1", "task": "detection" }, "input": { "resolution": "300x300", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 125, "latency_ms": 8, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack derived estimate — benchmark corpus", "notes": null, "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|ssd_mobilenet|v1|rknn|int8|300x300|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0238-rk3588-npu-ssd-mobilenet-v1-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "l", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 11.1, "latency_ms": 90.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip Electronics Co., Ltd.", "document": "Rockchip RKNN-Toolkit2 GitHub Repository — Benchmark Outputs", "doc_id": "bench:rockchip-electronics-co-ltd-rockchip-rknn-toolkit2-github-repository-benchmark-outputs", "url": "https://github.com/airockchip/rknn-toolkit2", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11|l|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0066-rk3588-npu-yolo11-l-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 16.9, "latency_ms": 59.2, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip Electronics Co., Ltd.", "document": "Rockchip RKNN-Toolkit2 GitHub Repository — Benchmark Outputs", "doc_id": "bench:rockchip-electronics-co-ltd-rockchip-rknn-toolkit2-github-repository-benchmark-outputs", "url": "https://github.com/airockchip/rknn-toolkit2", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11|m|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0065-rk3588-npu-yolo11-m-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 50, "latency_ms": 20, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip Electronics Co., Ltd.", "document": "Rockchip RKNN-Toolkit2 GitHub Repository — Benchmark Outputs", "doc_id": "bench:rockchip-electronics-co-ltd-rockchip-rknn-toolkit2-github-repository-benchmark-outputs", "url": "https://github.com/airockchip/rknn-toolkit2", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0063-rk3588-npu-yolo11-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 31.6, "latency_ms": 31.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip Electronics Co., Ltd.", "document": "Rockchip RKNN-Toolkit2 GitHub Repository — Benchmark Outputs", "doc_id": "bench:rockchip-electronics-co-ltd-rockchip-rknn-toolkit2-github-repository-benchmark-outputs", "url": "https://github.com/airockchip/rknn-toolkit2", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0064-rk3588-npu-yolo11-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11", "series": "yolo11", "variant": "x", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 6.4, "latency_ms": 156.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "low", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip Electronics Co., Ltd.", "document": "Rockchip RKNN-Toolkit2 GitHub Repository — Benchmark Outputs", "doc_id": "bench:rockchip-electronics-co-ltd-rockchip-rknn-toolkit2-github-repository-benchmark-outputs", "url": "https://github.com/airockchip/rknn-toolkit2", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11|x|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0067-rk3588-npu-yolo11-x-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 45.7, "latency_ms": 21.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "airockchip/rknn_model_zoo v2.3.2 — Official Benchmark Table", "doc_id": "bench:publisher-airockchip-rknn-model-zoo-v2-3-2-official-benchmark-table", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11_pose|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0375-rk3588-npu-yolo11-pose-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11_pose", "series": "yolo11_pose", "variant": "s", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 25.1, "latency_ms": 39.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "low", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "airockchip/rknn_model_zoo v2.3.2 — Official Benchmark Table", "doc_id": "bench:publisher-airockchip-rknn-model-zoo-v2-3-2-official-benchmark-table", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11_pose|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0376-rk3588-npu-yolo11-pose-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "m", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 9.9, "latency_ms": 101.3, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "airockchip/rknn_model_zoo v2.3.2 — Official Benchmark Table", "doc_id": "bench:publisher-airockchip-rknn-model-zoo-v2-3-2-official-benchmark-table", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11_seg|m|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0374-rk3588-npu-yolo11-seg-m-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "n", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 49.8, "latency_ms": 20.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "airockchip/rknn_model_zoo v2.3.2 — Official Benchmark Table", "doc_id": "bench:publisher-airockchip-rknn-model-zoo-v2-3-2-official-benchmark-table", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11_seg|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0372-rk3588-npu-yolo11-seg-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolo11_seg", "series": "yolo11_seg", "variant": "s", "task": "segmentation" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 25.1, "latency_ms": 39.9, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": "hw_only", "sources": [ { "publisher": "Publisher", "document": "airockchip/rknn_model_zoo v2.3.2 — Official Benchmark Table", "doc_id": "bench:publisher-airockchip-rknn-model-zoo-v2-3-2-official-benchmark-table", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolo11_seg|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0373-rk3588-npu-yolo11-seg-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov10", "series": "yolov10", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 50.8, "latency_ms": 19.7, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov10|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0242-rk3588-npu-yolov10-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 27.8, "latency_ms": 36, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov5|m|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0241-rk3588-npu-yolov5-m-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 68.5, "latency_ms": 14.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov5|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0239-rk3588-npu-yolov5-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov5", "series": "yolov5", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 46.5, "latency_ms": 21.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov5|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0240-rk3588-npu-yolov5-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "m", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 21.5, "latency_ms": 46.5, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov8|m|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0234-rk3588-npu-yolov8-m-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "n", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 61, "latency_ms": 16.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov8|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0232-rk3588-npu-yolov8-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov8", "series": "yolov8", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 36.5, "latency_ms": 27.4, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov8|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0233-rk3588-npu-yolov8-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolov8_pose", "series": "yolov8_pose", "variant": "n", "task": "pose" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 46.4, "latency_ms": 21.6, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolov8_pose|n|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0244-rk3588-npu-yolov8-pose-n-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588_npu", "power_mode": null, "software": null, "model": { "family": "yolox", "series": "yolox", "variant": "s", "task": "detection" }, "input": { "resolution": "640x640", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 35.6, "latency_ms": 28.1, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "C", "confidence": "medium", "date": "2026", "derivation": null, "notes": null, "sources": [ { "publisher": "Rockchip / airockchip (official GitHub org)", "document": "RKNN Model Zoo — Official Benchmark Table (README.md)", "doc_id": "bench:rockchip-airockchip-official-github-org-rknn-model-zoo-official-benchmark-table-readme-md", "url": "https://github.com/airockchip/rknn_model_zoo", "retrieved": "2026-09-16", "verified": "2026-09-16", "class": "C" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588_npu|yolox|s|rknn|int8|640x640|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0243-rk3588-npu-yolox-s-rknn-int8" }, { "origin": "catalog", "hardware": "rk3588s", "power_mode": null, "software": null, "model": { "family": "mobilenetv2", "series": "mobilenetv2", "variant": "v2", "task": "classification" }, "input": { "resolution": "224x224", "batch": 1 }, "runtime": "rknn", "precision": "int8", "accelerator": "npu", "scope": "inference_only", "metrics": { "fps": 450.7, "latency_ms": 2.22, "latency_p99_ms": null, "tokens_per_s": null, "power_w": null, "memory_mb": null, "accuracy_map50_95": null }, "class": "D", "confidence": "low", "date": "2026", "derivation": "EdgeAIStack cross-platform scaling — benchmark corpus", "notes": "Derived, not measured on an RK3588S: the RK3588 single-core RKNN Model Zoo figure (450.7 fps, zoo v2.3.2) carried at factor 1.0, because the RK3588S datasheet (V1.0, section 1.2.6) specifies the same NPU as the RK3588 datasheet — 6 TOPS, three cores, 384 KB × 3 internal buffer. Assumes the board holds the maximum NPU frequency the zoo measured at; sustained clocks on a passively cooled RK3588S board are not verified.", "sources": [ { "publisher": "EdgeAIStack", "document": "Benchmark corpus — research-derived rows, class D", "doc_id": "edgeaistack:benchmark-corpus", "url": "https://edgeaistack.ai/datasets/benchmark-corpus.json", "retrieved": "2026-09-28", "verified": "2026-09-28", "class": "D", "section": "benchmark corpus" } ], "power_condition": { "kind": "unspecified", "watts": null, "source_value": "default" }, "key": "rk3588s|mobilenetv2|v2|rknn|int8|224x224|1", "duplicates": [], "latency_source": "derived_reciprocal", "latency_p99_source": null, "latency_p99_basis": null, "id": "b0691-rk3588s-mobilenetv2-v2-rknn-int8" } ] }