Granite-Vision-4.1-4B on DGX Spark — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Granite-Vision-4.1-4B
Parameters	4.0B
Device	DGX Spark
Quantization	BF16
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	16
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.30	26.3	≤3 s
2	0.42	30.5	≤3 s
4	0.50	27.5	≤3 s
8	0.73	22.4	≤3 s
16	1.50	17.9	≤3 s
32	3.11	11.9	>3 s
64	6.68	7.6	>3 s
96	15.8	6.9	>3 s
104	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.30	26.3	≤3 s
2	0.57	26.0	≤3 s
4	1.16	27.9	≤3 s
8	2.39	25.0	≤3 s
16	5.28	23.4	>3 s
32	12.8	19.8	>3 s
64	—	—	Failed

Granite-Vision-4.1-4B (4.0B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.3 s, then writes 26.3 tok/s.

With 32 images per camera at 720p, the answer starts after 12.78 s. With one 2K image, after 0.36 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.