GLM-4.6V-Flash on DGX Spark — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	GLM-4.6V-Flash
Parameters	10B
Device	DGX Spark
Quantization	BF16
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	10.5	≤3 s
2	0.73	12.1	≤3 s
4	0.92	11.3	≤3 s
8	1.71	10.3	≤3 s
16	3.62	8.6	>3 s
32	5.99	6.0	>3 s
48	8.86	4.4	>3 s
52	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.60	10.5	≤3 s
2	1.24	10.5	≤3 s
4	2.43	10.4	≤3 s
8	4.97	10.3	>3 s
16	10.8	10.2	>3 s
32	25.1	9.9	>3 s
64	64.5	9.4	>3 s
128	—	—	Failed

GLM-4.6V-Flash (10B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.6 s, then writes 10.5 tok/s.

With 64 images per camera at 720p, the answer starts after 64.53 s. With one 2K image, after 2.81 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.