d1-3B on DGX Spark — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	d1-3B
Parameters	3.1B
Device	DGX Spark
Quantization	BF16
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	64
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.18	34.5	≤3 s
2	0.25	40.5	≤3 s
4	0.26	36.8	≤3 s
8	0.41	33.8	≤3 s
16	0.78	27.2	≤3 s
32	1.33	20.9	≤3 s
64	2.68	14.1	≤3 s
128	7.36	10.2	>3 s
256	20.3	8.7	>3 s
288	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.18	34.5	≤3 s
2	0.32	34.3	≤3 s
4	0.61	34.8	≤3 s
8	1.20	33.9	≤3 s
16	2.43	33.8	≤3 s
32	4.97	34.0	>3 s
64	—	—	Failed

d1-3B (3.1B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.18 s, then writes 34.5 tok/s.

With 32 images per camera at 720p, the answer starts after 4.97 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 64 cameras at once.