Benchmark Explorer · YOLOv8n on Hailo-8L

YOLOv8n on Hailo-8L: 2 rows.

Method v1.4.1 · dataset 2026-09-29 · verified 2026-09-16 · last updated September 2026

yolov8n on Hailo-8L: 2 rows (2 measured, 0 derived). Measured share for this scope: 67% of 46 rows (class C). Expected: INT8 202–438 fps. Fastest: Hailo-8 1036 fps (class C), NVIDIA A100 (server) 1010.1 fps (class C), Jetson Thor T5000 859 fps (class D).

Change hardware, precision, runtime or add a comparison set on the live engine.

Expected range per precision

Min / median / max of the matching rows, all classes. When a precision has no row, the estimator's benchmark hierarchy (B → C → D) supplies a fallback with its tier chain reported.

Precision	Metric	N	Min	Median	Max	Best class	Measured
INT8	fps	2	202	320	438	C	2
range pools 2 rows across batch (FPS (Batch Size=1), FPS (Batch Size=8)); not one measurement condition

Rows

Precision	Runtime	Input	Power mode	Software	Metric	Class	Date	Source	Derivation	Dupes
INT8	hailo sdk	640x640 b8	—	hailo_dataflow_compiler v2.19.0	438 fps	C	2026	Hailo · bench:hailo-hailo-model-zoo-object-detection-benchmarks · verified 2026-09-16		2
INT8	hailo sdk	640x640 b1	—	hailo_dataflow_compiler v2.19.0	202 fps	C	2026	Hailo · bench:hailo-hailo-model-zoo-object-detection-benchmarks · verified 2026-09-16		2

Cross-hardware comparison

Best row per hardware for the same model and precision, relative to the fastest.

Hardware	Best value	Class
Hailo-8	1036 fps	C
NVIDIA A100 (server) (reference)	1010.1 fps	C
Jetson Thor T5000	859 fps	D
NVIDIA L4 (24GB)	816 fps	D
NVIDIA A10 (24GB)	594 fps	D
Jetson AGX Orin 64GB	504 fps	C
Hailo-8L	438 fps	C
Hailo-10H	375 fps	C
Jetson AGX Orin 32GB	352 fps	D
NVIDIA A2 (16GB)	352 fps	D
Jetson Orin Nano Super	341 fps	C
Jetson Orin NX 16GB	264 fps	C
Jetson Orin Nano 8GB	245 fps	D
Jetson Orin NX 8GB	160 fps	D
Rockchip RK3588 NPU	61 fps	C
Jetson Orin Nano 4GB (reference)	48 fps	C
Neousys Nuvo-9531 (Intel 12th-14th Gen Core)	28 fps	C
Raspberry Pi 5 (CPU) (reference)	12.4 fps	D

Assumptions

Latency provenance: 2 of the 2 matching rows that report a latency computed it as 1000/fps rather than measuring it (latency_source → derived_reciprocal); 0 were observed independently. Every p99 in the database is a modelled 1.20–1.43× tail over the p50, never an observed one.
Class C = measured and published by the source named on the row; class D = EdgeAIStack-derived (derivation stated). Rows are inference-only unless scope says end_to_end; publisher measurement conditions (JetPack, batch, warm-up) are in the linked page.

Method and limitations

Rows are read from the normalised benchmark database (publisher benchmarks and research-derived rows, deduplicated per hardware × model × runtime × precision × input × power mode, best evidence kept). Expected range = min / median / max of the matching rows per precision; when a precision has no row the estimator's benchmark hierarchy (B → C → D) is consulted and its tier chain reported. Comparison = best row per hardware for the same model and precision, relative to the fastest. Nothing is scored; every number carries its class and source line.

Full method, dedupe rule, evidence classes and the confidence rule: Benchmark Explorer methodology.

Measure it yourself.

Nothing measured for your exact config, or want to confirm a class-D row? Run the measurement protocol on your board and submit the result — a human checks it against the protocol before it becomes a class-C row with your attribution.

SUBMIT A MEASUREMENT →

Change one input and re-run.

The live engine keeps every source line and gives you a fresh permalink and summary.

OPEN THIS QUERY LIVE →