Llama-3.1-Nemotron-Nano-VL-8B on Jetson AGX Thor — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Llama-3.1-Nemotron-Nano-VL-8B
Parameters	8.7B
Device	Jetson AGX Thor
Quantization	BF16
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	8
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.75	15.3	≤3 s
2	0.81	14.0	≤3 s
4	1.10	12.6	≤3 s
8	1.84	10.4	≤3 s
16	3.90	8.1	>3 s
32	6.83	4.9	>3 s
40	8.83	4.1	>3 s
44	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.75	15.3	≤3 s
2	1.49	15.0	≤3 s
4	3.18	14.5	>3 s
8	7.37	13.6	>3 s
16	19.1	12.2	>3 s
32	55.7	9.8	>3 s
64	—	—	Failed

Llama-3.1-Nemotron-Nano-VL-8B (8.7B parameters), served in BF16 format with vLLM on Jetson AGX Thor, starts answering a single camera that sends one 720p image after 0.75 s, then writes 15.3 tok/s.

With 32 images per camera at 720p, the answer starts after 55.68 s. With one 2K image, after 0.84 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 8 cameras at once.