inference/research
verifiedmedium · 0.80updated 2026-10-17

NVIDIA Developer Jetson Benchmarks — Jetson AGX Thor vLLM LLM table (ISL/OSL 2048/128)

benchmark

Fields

id
sources/nvidia-jetson-benchmarks-thor-vllm-table
Publisher
NVIDIA
Kind
benchmark
Accessed
2026-10-17
notes
Vendor page, section 'NVIDIA Jetson AGX Thor Benchmarking Results'. Table gives tokens/sec at max concurrency 1 and max concurrency 8 for six LLMs. Footnote, quoted: 'These results were achieved with the NVIDIA Jetson AGX Thor Developer Kit running NVIDIA JetPack 7.0, NVIDIA CUDA 13.0, and NVIDIA TensorRT 10.13. LLM and VLM benchmarks were performed using VLLM with ISL/OSL as 2048/128.' Weight dtype and vLLM version are not stated. Page read 2026-10-17; the page itself carries no date. VLM rows on the same page are out of scope for this repo's text-inference records and were not filed.

Notes

Vendor page, section 'NVIDIA Jetson AGX Thor Benchmarking Results'. Table gives tokens/sec at max concurrency 1 and max concurrency 8 for six LLMs. Footnote, quoted: 'These results were achieved with the NVIDIA Jetson AGX Thor Developer Kit running NVIDIA JetPack 7.0, NVIDIA CUDA 13.0, and NVIDIA TensorRT 10.13. LLM and VLM benchmarks were performed using VLLM with ISL/OSL as 2048/128.' Weight dtype and vLLM version are not stated. Page read 2026-10-17; the page itself carries no date. VLM rows on the same page are out of scope for this repo's text-inference records and were not filed.

Cross-references

This record references no other records.