NVIDIA Jetson AGX Orin 64GB — Benchmarks & Specs

Bottom line: how fast is the NVIDIA Jetson AGX Orin 64GB?

SpecPicks median: 28.0 tokens/sec across 3 cited single-user runs of 7-9B dense models at 4-bit (Q4) under llama.cpp (range 1.5–31.0 tok/s), from MultimodalFlow (2 runs), DFRobot. At the $3,379.00 Amazon price of the cheapest linked listing (checked 2026-10-22 16:17 UTC; prices change), that is a SpecPicks Value Index of 8.29 tok/s per $1,000. For local LLM inference it generates 27.0 tokens/sec running qwen3:30b-a3b (a mixture-of-experts model with ~3B active parameters, so dense models of similar size run far slower) at q4_K_M under llama.cpp, per Hackster.io. In MLPerf v3.1 ResNet-50 Image Classification Offline it scores 6,423.63 samples/s, per NVIDIA Developer. It has 64 GB of memory shared by the CPU and GPU, and the qwen3:30b-a3b run above (30B parameters) is the largest model on file at 4-bit quantization on it. Larger models have no sourced run on file.

Every figure above is a row in the tables below, and each row links out to the review or public benchmark database the number was taken from; the memory size is the on-paper spec; the SpecPicks median is computed here from the runs it names, each a row in the AI table, and the per-$1,000 figure divides it by the dated listing price. SpecPicks aggregates published measurements; it does not report first-party benchmark runs.