VLM Inference Benchmark Explorer
Gemma-4-26B-A4B-it on Jetson AGX Orin 32GB — Q4_0, llama.cpp vision-language inference benchmark
Open this configuration in the VLM Inference Benchmark Explorer
| Model | Gemma-4-26B-A4B-it |
|---|---|
| Parameters | 26B |
| Device | Jetson AGX Orin 32GB |
| Quantization | Q4_0 |
| Image size | 720p |
| Images per camera | 1 |
| Max Cameras (TTFT ≤ 3000 ms) | >3000 ms |
| Engine | llama.cpp |
| Cameras | TTFT (ms) | TPS (tok/s) | Target |
|---|---|---|---|
| 1 | 3587 | 25.1 | >3000 ms |
| 2 | 4260 | 9.4 | >3000 ms |
| 4 | 9657 | 7.0 | >3000 ms |
| 8 | 23901 | 9.4 | >3000 ms |
| Images per request | TTFT (ms) | TPS (tok/s) | Target |
|---|---|---|---|
| 1 | 3587 | 25.1 | >3000 ms |
| 3 | 15219 | 24.4 | >3000 ms |
| 5 | 24097 | 23.9 | >3000 ms |
Gemma-4-26B-A4B-it (26B parameters), served in Q4_0 format with llama.cpp on Jetson AGX Orin 32GB, starts answering a single camera that sends one 720p image after 3,587 ms (time to first token, TTFT), then writes 25.1 tok/s.
With 5 images per camera at 720p, the answer starts after 24,097 ms. With one 2K image, after 15,438 ms.
At the default target of a mean TTFT within 3,000 ms (720p, one image per camera), not even one camera is answered in time.
Gemma-4-26B-A4B-it (26B parameters), served in Q4_0 format with llama.cpp on Jetson AGX Orin 32GB, starts answering a single camera that sends one 720p image after 3,587 ms (time to first token, TTFT), then writes 25.1 tok/s.
With 5 images per camera at 720p, the answer starts after 24,097 ms. With one 2K image, after 15,438 ms.
At the default target of a mean TTFT within 3,000 ms (720p, one image per camera), not even one camera is answered in time.