Qwen3.5-2B on Jetson Orin NX 16GB — Q8_0, llama.cpp vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Qwen3.5-2B
Parameters	2.3B
Device	Jetson Orin NX 16GB
Quantization	Q8_0
Image size	720p
Images per camera	1
Max Cameras (response ≤ 3 s)	1
Engine	llama.cpp

Cameras	Response (s)	TPS (tok/s)	Target
1	2.09	25.5	≤3 s
2	3.26	25.6	>3 s
4	11.8	29.8	>3 s
8	24.6	30.4	>3 s

Images per request	Response (s)	TPS (tok/s)	Target
1	2.09	25.5	≤3 s
2	4.11	25.3	>3 s
4	8.12	25.0	>3 s
8	16.2	24.3	>3 s
16	32.5	23.1	>3 s
32	66.5	21.1	>3 s

Qwen3.5-2B (2.3B parameters), served in Q8_0 format with llama.cpp on Jetson Orin NX 16GB, starts answering a single camera that sends one 720p image after 2.09 s, then writes 25.5 tok/s.

With 32 images per camera at 720p, the answer starts after 66.47 s. With one 2K image, after 15.23 s.

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