VLM Inference Benchmark Explorer
Gemma-4-31B-it on DGX Spark — BF16, vLLM vision-language inference benchmark
Open this configuration in the VLM Inference Benchmark Explorer
| Model | Gemma-4-31B-it |
|---|---|
| Parameters | 31B |
| Device | DGX Spark |
| 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 | 0.78 | 3.8 | ≤3 s |
| 2 | 1.11 | 3.7 | ≤3 s |
| 4 | 1.31 | 3.6 | ≤3 s |
| 8 | 1.92 | 3.4 | ≤3 s |
| 16 | 4.00 | 3.3 | >3 s |
| 32 | 5.44 | 2.6 | >3 s |
| 64 | 9.47 | 2.0 | >3 s |
| 72 | — | — | Failed |
| Images per request | Response (s) | TPS (tok/s) | Target |
|---|---|---|---|
| 1 | 0.78 | 3.8 | ≤3 s |
| 2 | 1.09 | 3.8 | ≤3 s |
| 4 | 1.98 | 3.8 | ≤3 s |
| 8 | 3.99 | 3.8 | >3 s |
| 16 | 8.77 | 3.7 | >3 s |
| 32 | 20.7 | 3.7 | >3 s |
| 64 | 57.8 | 3.7 | >3 s |
| 128 | 228.0 | 3.6 | >3 s |
| 256 | — | — | Failed |
Gemma-4-31B-it (31B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.78 s, then writes 3.8 tok/s.
With 128 images per camera at 720p, the answer starts after 228.04 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.
Gemma-4-31B-it (31B parameters), served in BF16 format with vLLM on DGX Spark, starts answering a single camera that sends one 720p image after 0.78 s, then writes 3.8 tok/s.
With 128 images per camera at 720p, the answer starts after 228.04 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.