Eagle2.5-8B on Jetson AGX Thor — BF16, vLLM vision-language inference benchmark

Open this configuration in the VLM Inference Benchmark Explorer

Model	Eagle2.5-8B
Parameters	8.1B
Device	Jetson AGX Thor
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	0.65	16.9	≤3 s
2	0.86	16.4	≤3 s
4	1.04	10.2	≤3 s
8	1.54	11.2	≤3 s
16	4.60	5.3	>3 s
32	11.0	4.7	>3 s
48	17.7	4.3	>3 s
52	20.1	17.4	>3 s
56	—	—	Failed

Images per request	Response (s)	TPS (tok/s)	Target
1	0.65	16.9	≤3 s
2	1.27	17.3	≤3 s
4	2.67	16.0	≤3 s
8	6.12	17.2	>3 s
16	15.6	14.6	>3 s
32	44.4	13.4	>3 s
64	143.7	10.7	>3 s
128	—	—	Failed

Eagle2.5-8B (8.1B parameters), served in BF16 format with vLLM on Jetson AGX Thor, starts answering a single camera that sends one 720p image after 0.65 s, then writes 16.9 tok/s.

With 64 images per camera at 720p, the answer starts after 143.68 s. With one 2K image, after 0.74 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.