Qwen3.6-35B-A3B on Jetson AGX Thor — FP8, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Qwen3.6-35B-A3B
Parameters	35B
Device	Jetson AGX Thor
Quantization	FP8
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	8
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.49	53.7	≤3 s
2	0.77	43.5	≤3 s
4	1.22	32.5	≤3 s
8	1.80	21.0	≤3 s
16	3.07	14.1	>3 s
32	3.97	8.9	>3 s
64	6.47	5.4	>3 s
96	9.14	3.9	>3 s
104	9.86	3.6	>3 s
112	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.49	53.7	≤3 s
2	0.71	53.5	≤3 s
4	1.35	53.2	≤3 s
8	2.61	52.3	≤3 s
16	5.32	51.0	>3 s
32	11.4	48.2	>3 s
64	27.0	44.0	>3 s
128	63.8	36.8	>3 s
256	178.1	27.5	>3 s
512	—	—	Failed

Qwen3.6-35B-A3B (35B parameters), served in FP8 format with vLLM on Jetson AGX Thor, starts answering a single camera that sends one 720p image after 0.49 s, then writes 53.7 tok/s.

With 256 images per camera at 720p, the answer starts after 178.09 s. With one 2K image, after 2.04 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.