Gemma-4-31B-it on DGX Spark — NVFP4, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Gemma-4-31B-it
Parameters	31B
Device	DGX Spark
Quantization	NVFP4
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	16
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.51	7.1	≤3 s
2	0.71	7.0	≤3 s
4	0.80	6.7	≤3 s
8	1.36	6.4	≤3 s
16	2.64	6.0	≤3 s
32	3.96	4.8	>3 s
64	6.72	3.5	>3 s
96	9.30	2.7	>3 s
104	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.51	7.1	≤3 s
2	0.81	7.0	≤3 s
4	1.53	7.0	≤3 s
8	3.24	7.0	>3 s
16	6.81	6.9	>3 s
32	16.6	6.9	>3 s
64	57.3	6.8	>3 s
128	208.3	6.6	>3 s
256	—	—	Failed

Gemma-4-31B-it (31B parameters), served in NVFP4 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.51 s, then writes 7.1 tok/s.

With 128 images per camera at 720p, the answer starts after 208.31 s. With one 2K image, after 0.54 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 16 cameras at once.