← sources

Nemotron 3.5 Bolt E4B on Jetson Orin Nano Super at 27.7 tok/s!

source: reddit (reddit:LocalLLaMA)published 2026-11-05 17:20ingested 2026-11-06 02:39

retained text

Hi! I wanted to share that we just got NVIDIA's latest Nemotron Bolt model running at 27.7 tok/s (compared to 15 tok/s on the official release with Llama CPP) on Jetson Orin Nano Super (1.8x faster)! It can program games on a model running completely on a cheap edge device!

We have our own Jetson devices serving that model on our website, if you want to get a feel for how the model runs on the device here or run it locally as a TensorRT engine with our own variant of TensorRT edge LLM supporting int4 quantization:

[https://embedl.com/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16/orin-nano-super?tab=run](https://embedl.com/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16/orin-nano-super?tab=run)

document fetched 2026-11-06 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1xoogj8/nemotron_35_bolt_e4b_on_jetson_orin_nano_super_at