Qwen3.5-122B-A10B on DGX Spark — NVFP4, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Qwen3.5-122B-A10B
Parameters	122B
Device	DGX Spark
Quantization	NVFP4
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	8
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.60	14.2	≤3 s
2	1.05	14.3	≤3 s
4	1.69	11.6	≤3 s
8	2.45	8.5	≤3 s
16	3.87	5.9	>3 s
20	4.43	5.2	>3 s
24	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.60	14.2	≤3 s
2	0.95	14.2	≤3 s
4	1.90	14.2	≤3 s
8	3.83	14.2	>3 s
16	7.80	14.0	>3 s
32	17.1	13.8	>3 s
64	40.0	13.5	>3 s
128	103.3	13.1	>3 s
256	—	—	Failed

Qwen3.5-122B-A10B (122B parameters), served in NVFP4 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.6 s, then writes 14.2 tok/s.

With 128 images per camera at 720p, the answer starts after 103.26 s. With one 2K image, after 2.22 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.