Can an Orin NX 16GB decode 4 × 1440p H.265 at 20 fps? Yes: 27 % of the rated NVDEC throughput, in every power mode.

Answer

Four 2560×1440 streams at 20 fps are 294.9 MP/s. NVIDIA rates the Orin NX H.265 decoder at 1,100 MP/s (18 × 1080p30, 4 × 4K30), so the request uses 27 % of the decoder: PASS with wide margin. 1440p20 is not a printed datasheet point, so the number is a pixel-rate comparison (class D) against the class A table; NVIDIA's own reply used the same reasoning by bracketing 1440p20 under the 4K30 figure.

The power-mode half of the question is settled by NVIDIA staff in the same thread: the NVDEC maximum clock is the same in every nvpmodel preset, so 25 W and MAXN SUPER decode identically. Power mode changes GPU and CPU clocks, which affects the inference row below, not decode. Verify at runtime with sudo tegrastats and watch the NVDEC line.

The question

• 2026-08-16 · thread 380328

  “Is it correct to treat 4× 2560×1440 at 20fps as comfortably within the 4× 4K30 figure? … Is decode throughput the same at 25W as at the 40W MAXN SUPER profile?”

• 2026-08-17 · reply by NVIDIA staff (DaneLLL)

  “Because it can do 4x 3840x2160p30 H265 decoding, it is supposed to work well for 4x [2560×1440]. Maximum frequency of NVDEC is same in different power modes.”

Every stage as a constraint

Required against available in the same unit; PASS below 80 %, NEAR_LIMIT from 80 %, FAIL at 100 %. Evidence class and confidence per row; the source line is the document the available figure came from.

Capacity landmarks for this workload: 8 cameras comfortable · 10 near limit · 10 at the modeled ceiling (D-class cross_platform_scaled) · fails at 11. Overall confidence LOW (weakest stage).

Assumptions the engine used

• 2560×1440@20 is not a published datasheet point; capacity is interpolated from the 1100 MP/s throughput NVIDIA rates for H265 (Main, Main10).
• Evidence age assessed as of 2026-09-07; decode confidence drops one step after 365 days and two after 730 days since verification.
• Single-stream modelled 176.2 fps (cross_platform_scaled); multi-stream efficiency 0.88 at 4 streams.
• Power mode 40w: ×1.15 of the row's throughput = calibration 1.15 at 40w ÷ 1 at the reference mode (the row states no power mode) (EdgeAIStack calibration, class D).
• Hardware (VIC/GPU) preprocessing budget ≈480 detection frames/s per node (engineering heuristic).
• Shared memory budget: weights 11.2 MB + activations 41.4 MB + builder workspace 64 MB + io 0 MB + decode buffers 64 MB + preprocess 128 MB + tracker 0 MB + OS 900 MB + 10% reserve.
• Memory evidence class E: fallback or interpolated model parts.
• 4 × ~2.67 Mbit/s average (4.27 peak) H265 at 1440p@20 from the camera bitrate table; module ingress sized to 1 GbE MAC capability on Orin NX 16GB (Super, 40W).
• Production Ethernet ingress depends on the selected carrier PHY; module MAC capability is used until a carrier is confirmed.

Reproduce and cite

Live permalink (carries the method and dataset versions; the page flags if either has changed since):
https://edgeaistack.ai/engines/camera-stream-capacity/?hw=jetson_orin_nx&res=2560x1440&fps=20&mv=1.8.0&dv=2026-09-07

Static page: https://edgeaistack.ai/decode/answers/orin-nx-16gb-4x-1440p20-h265/

Sources

• NVIDIA · DS-10712-001 v1.7 §2.10.1 Multi-Standard Video Decoder, Table 2-7: Supported Video Decode Streams · verified 2026-09-13 · class D

  Jetson Orin NX (16GB / 8GB) H265 decode: 1 × 8K30, 2 × 4K60, 4 × 4K30, 9 × 1080p60, 18 × 1080p30 (1100 MP/s)

• D · EdgeAIStack model (cross_platform_scaled) · class D

  yolov8s int8 tensorrt on jetson_orin_nx: 176.2 fps single stream

• NVIDIA · developer.download.nvidia.com/assets/embedded/secure/jetson/orin_nx/docs/Jetson-Orin-NX-Series-Modules-Datashe · verified 2026-09-08 · class A

  NVIDIA Jetson Orin NX 16GB: 1 × 1 GbE MAC (module; the carrier decides the PHY count)

• NVIDIA Developer Forums, thread 380328 · reply by NVIDIA staff (DaneLLL), 2026-08-17 · class A

Method: Camera Stream Capacity methodology v1.8.0. Module decode tables: Jetson Orin NX (16GB / 8GB) H.265.