VLM Inference Benchmark Explorer
Llama-4-Scout-17B-16E-Instruct on DGX Spark — NVFP4, vLLM vision-language inference benchmark
Open this configuration in the VLM Inference Benchmark Explorer
| Model | Llama-4-Scout-17B-16E-Instruct |
|---|---|
| Parameters | 109B |
| Device | DGX Spark |
| Quantization | NVFP4 |
| Image size | 720p |
| Images per camera | 1 |
| Max Cameras (response ≤ 3 s) | 8 |
| Engine | vLLM |
| Cameras | Response (s) | TPS (tok/s) | Target |
|---|---|---|---|
| 1 | 0.94 | 18.4 | ≤3 s |
| 2 | 1.33 | 14.7 | ≤3 s |
| 4 | 1.54 | 10.1 | ≤3 s |
| 8 | 2.56 | 7.0 | ≤3 s |
| 16 | 5.02 | 4.8 | >3 s |
| 24 | 7.34 | 3.9 | >3 s |
| 28 | 8.44 | 3.4 | >3 s |
| 32 | — | — | Failed |
| Images per request | Response (s) | TPS (tok/s) | Target |
|---|---|---|---|
| 1 | 0.94 | 18.4 | ≤3 s |
| 2 | 1.89 | 18.1 | ≤3 s |
| 4 | 3.93 | 17.6 | >3 s |
| 8 | 8.03 | 17.4 | >3 s |
| 16 | 17.0 | 17.1 | >3 s |
| 32 | 37.3 | 16.4 | >3 s |
| 64 | 89.1 | 14.4 | >3 s |
| 128 | — | — | Failed |
Llama-4-Scout-17B-16E-Instruct (109B parameters), served in NVFP4 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.94 s, then writes 18.4 tok/s.
With 64 images per camera at 720p, the answer starts after 89.06 s. With one 2K image, after 1.39 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.
Llama-4-Scout-17B-16E-Instruct (109B parameters), served in NVFP4 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.94 s, then writes 18.4 tok/s.
With 64 images per camera at 720p, the answer starts after 89.06 s. With one 2K image, after 1.39 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.