NVIDIA-Nemotron-3.5-Super-VL-120B-A12B on DGX Spark — NVFP4, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	NVIDIA-Nemotron-3.5-Super-VL-120B-A12B
Parameters	120B
Device	DGX Spark
Quantization	NVFP4
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	4
Engine	vLLM

Cameras	Response (s)	TPS (tok/s)	Target
1	0.77	19.1	≤3 s
2	1.31	15.7	≤3 s
4	2.03	12.3	≤3 s
8	4.21	9.1	>3 s
16	7.44	6.3	>3 s
32	11.6	4.2	>3 s
48	14.6	3.1	>3 s
56	16.0	2.8	>3 s
60	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.77	19.1	≤3 s
2	1.08	18.9	≤3 s
4	1.88	18.9	≤3 s
8	4.00	18.9	>3 s
16	8.15	18.9	>3 s
32	21.8	18.8	>3 s
64	34.8	19.7	>3 s
128	—	—	Failed

NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (120B parameters), served in NVFP4 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.77 s, then writes 19.1 tok/s.

With 64 images per camera at 720p, the answer starts after 34.83 s. With one 2K image, after 2.46 s.

At the default target of an answer starting within 3 s (720p, one image per camera), it keeps up with 4 cameras at once.