// Platform

Jetson Orin NX 8GB for Edge AI — Specs, Sizing & Deployment Fit

Last updated: August 2026

The value cut of the NX: 117 TOPS on the same silicon as the 16GB (Ampere GPU, dual NVDLA, same ~102 GB/s bandwidth) with half the memory — enough for 4–8 detection streams per node at ~$649.

117 INT8 TOPS
10–40 W
~8 streams (est.)
~$649 module

Planning Takeaway

The value cut of the NX: 117 TOPS on the same silicon as the 16GB with half the memory, carrying 4–8 detection streams per node at ~$649. Because compute-per-dollar stays high, EdgeAIStack’s multi-node planner actively recommends it as the per-node building block for multi-camera 4K deployments — the constraint to watch is the 8 GB memory ceiling, not TOPS.

Specifications

SpecValue
Compute117 INT8 TOPS (Ampere GPU + 2× NVDLA)
Memory8 GB LPDDR5, ~102 GB/s
Power modes10W · 15W · MAXN · 40W Super · MAXN Super (10–40 W envelope)
Video decode (modeled planning ceiling)~10× 1080p / ~2× 4K H.264 concurrent
Est. planning capacity~8 concurrent 1080p detection streams
CoolingActive required
Form factorJetson module (69.6×45 mm)
Runtimes / precisionsTensorRT, PyTorch, ONNX, llama.cpp · FP32/FP16/INT8
SoftwareJetPack 7.2 (Ubuntu 24.04, CUDA 13, June 2026) — current; JetPack 6.x (Ubuntu 22.04) — legacy track
Indicative price~$649 (module, as of mid-2026; NX 16GB module ~$999)

Sizing Fit

Fits 4–8 camera 1080p detection nodes where 8 GB of memory covers the pipeline, and multi-node 4K deployments where its compute-per-dollar makes it the planner’s preferred per-node building block.

Decision Framework

Choose the Orin NX 8GB if:

  • You need 4–8 concurrent 1080p detection streams per node and your models fit in 8 GB.
  • You are scaling out — its compute-per-dollar makes it a strong per-node building block, and the multi-node planner recommends it for multi-camera 4K workloads.
  • You want DLA engines for power-efficient offload alongside the GPU without paying for the 16GB.

Choose a sibling instead if:

  • Jetson Orin Nano Super — your node serves ≤6 streams and ~$399 buys enough compute.
  • Jetson Orin NX 16GB — multi-model pipelines or tracking push past 8 GB of memory, or you need ~12 streams on one node.
  • Hailo-10H — detection-only at up to ~8 streams on an existing host at a fraction of the power (~$130).

Frequently Asked Questions

How much does the Jetson Orin NX 8GB cost?

The Orin NX 8GB module lists at ~$649 as of the July 2026 NVIDIA repricing (up from $399); the 16 GB variant lists at ~$999. Carrier board and cooling are additional.

How many cameras can the Orin NX 8GB handle?

EdgeAIStack’s planning model puts it around 8 concurrent 1080p detection streams at 15 FPS, with hardware decode ceilings near ~10× 1080p or ~2× 4K H.264. Real capacity depends on model, resolution, and pipeline — validate with the Hardware Selector.

Orin NX 8GB vs 16GB — what do you actually give up?

Memory and some clocks: 8 GB vs 16 GB, 117 vs 157 INT8 TOPS, at the same ~102 GB/s bandwidth on the same silicon (Ampere GPU, 2× NVDLA, one NVDEC). EdgeAIStack models it at ~8 streams vs ~12 for the 16GB. At ~$649 vs ~$999 the 8GB wins when your models fit its memory; step up when they don’t.

What JetPack version does the Orin NX 8GB run?

JetPack 7.2 (June 2026) is current for this module — Ubuntu 24.04, Linux kernel 6.8, CUDA 13. JetPack 6.x remains a widely deployed legacy track on Ubuntu 22.04. Moving an existing JetPack 6 install to 7 requires the OTA package or a full reflash — a plain apt upgrade is not an upgrade path.