Speed test matrix
gemma-4-12B-it on H200 NVL 141GB
vllm · 0.26.1rc1.dev638+g52be12cfa · BF166 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K
Hardware
- Type
- Discrete GPU
- GPU
- H200 NVL
- VRAM
- 141 GB
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.26.1rc1.dev638+g52be12cfa
- Quantization
- BF16
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
- Commit
52be12cfac0c
Run
- Served model
- gemma
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.4.29
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 30 s
- Max output
- 4,096 tokens
Launch command
"$HOME/env_h200-gemma-vllm/bin/vllm" serve google/gemma-4-12B-it --served-model-name gemma --trust-remote-code --tensor-parallel-size 1 --reasoning-parser gemma4 --enable-auto-tool-choice --tool-call-parser gemma4 --max-model-len 262144 --gpu-memory-utilization 0.849406 --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context19.4K422 ms TTFT
16K context18.9K860 ms TTFT
32K context16.9K1,919 ms TTFT
64K context13.6K4,745 ms TTFT
128K context9,73413,251 ms TTFT
251K context6,28840,199 ms TTFT
256K context6,21341,501 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 | 32 | 64 | 128 |
|---|---|---|---|---|---|---|---|---|
| 8K | 100.071 ms TTFT | 197.6100 ms TTFT | 382.3126 ms TTFT | 727.3187 ms TTFT | 1,323.8265 ms TTFT | 2,183.4391 ms TTFT | 3,247.7651 ms TTFT | Over capacity |
| 16K | 99.595 ms TTFT | 196.7136 ms TTFT | 378.1130 ms TTFT | 712.0262 ms TTFT | 1,278.7395 ms TTFT | 2,068.5618 ms TTFT | 2,972.1884 ms TTFT | Over capacity |
| 32K | 98.6134 ms TTFT | 193.6214 ms TTFT | 369.7286 ms TTFT | 683.7380 ms TTFT | 1,191.8685 ms TTFT | 1,849.91,019 ms TTFT | Over capacity | Over capacity |
| 64K | 97.2223 ms TTFT | 189.4324 ms TTFT | 355.6422 ms TTFT | 635.4653 ms TTFT | 1,059.01,200 ms TTFT | Over capacity | Over capacity | Over capacity |
| 128K | 94.5445 ms TTFT | 181.5616 ms TTFT | 330.0874 ms TTFT | 558.61,451 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 251K | 90.01,191 ms TTFT | 168.11,470 ms TTFT | 289.42,202 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.
TTFT is measured after the benchmark warms the server's prompt cache, so it mostly excludes prefill.
All measurements
Latencies are medians. Total tok/s is output across all simultaneous requests.| Phase | Context | Requests at once | Status | Total tok/s | Per request tok/s | TTFT | Time between tokens | Average active requests | Completed requests | Output tokens | Measured for |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Prefill | 8K | 1 | Valid | 19.4K | — | 422 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 18.9K | — | 860 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 16.9K | — | 1,919 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 13.6K | — | 4,745 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 9,734.0 | — | 13,251 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 6,288.0 | — | 40,199 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 6,213.0 | — | 41,501 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 100.0 | 100.1 | 71 ms | 10.0 ms | 1.0 | 0 | 3,001 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 197.6 | 98.9 | 100 ms | 10.1 ms | 2.0 | 0 | 5,929 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 382.3 | 95.7 | 126 ms | 10.5 ms | 4.0 | 0 | 11,469 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 727.3 | 91.1 | 187 ms | 11.0 ms | 8.0 | 0 | 21,816 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,323.8 | 83.0 | 265 ms | 12.1 ms | 16.0 | 0 | 39,712 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 2,183.4 | 68.4 | 391 ms | 14.6 ms | 32.0 | 0 | 65,505 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 3,247.7 | 51.0 | 651 ms | 19.6 ms | 64.0 | 0 | 97,401 | 30.0 s |
| Sustained decode | 8K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 99.5 | 99.6 | 95 ms | 10.0 ms | 1.0 | 0 | 2,985 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 196.7 | 98.5 | 136 ms | 10.2 ms | 2.0 | 0 | 5,901 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 378.1 | 94.6 | 130 ms | 10.6 ms | 4.0 | 0 | 11,348 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 712.0 | 89.2 | 262 ms | 11.2 ms | 8.0 | 0 | 21,360 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,278.7 | 80.2 | 395 ms | 12.5 ms | 16.0 | 0 | 38,352 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 2,068.5 | 64.8 | 618 ms | 15.4 ms | 32.0 | 0 | 62,050 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 2,972.1 | 46.7 | 884 ms | 21.4 ms | 64.0 | 0 | 89,160 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 98.6 | 98.7 | 134 ms | 10.1 ms | 1.0 | 0 | 2,958 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 193.6 | 96.9 | 214 ms | 10.3 ms | 2.0 | 0 | 5,812 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 369.7 | 92.6 | 286 ms | 10.8 ms | 4.0 | 0 | 11,088 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 683.7 | 85.6 | 380 ms | 11.7 ms | 8.0 | 0 | 20,512 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,191.8 | 74.4 | 685 ms | 13.4 ms | 16.0 | 0 | 35,744 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 1,849.9 | 57.6 | 1,019 ms | 17.4 ms | 32.0 | 0 | 55,488 | 30.0 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 97.2 | 97.4 | 223 ms | 10.3 ms | 1.0 | 0 | 2,914 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 189.4 | 94.8 | 324 ms | 10.5 ms | 2.0 | 0 | 5,680 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 355.6 | 89.0 | 422 ms | 11.2 ms | 4.0 | 0 | 10,668 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 635.4 | 79.2 | 653 ms | 12.6 ms | 8.0 | 0 | 19,068 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 1,059.0 | 65.8 | 1,200 ms | 15.2 ms | 16.0 | 0 | 31,758 | 30.0 s |
| Sustained decode | 64K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 94.5 | 94.8 | 445 ms | 10.5 ms | 1.0 | 0 | 2,837 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 181.5 | 91.0 | 616 ms | 11.0 ms | 2.0 | 0 | 5,448 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 330.0 | 82.5 | 874 ms | 12.1 ms | 4.0 | 0 | 9,900 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 558.6 | 69.6 | 1,451 ms | 14.4 ms | 8.0 | 0 | 16,752 | 30.0 s |
| Sustained decode | 128K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 1 | Valid | 90.0 | 90.4 | 1,191 ms | 11.1 ms | 1.0 | 0 | 2,700 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 168.1 | 84.4 | 1,470 ms | 11.8 ms | 2.0 | 0 | 5,042 | 30.0 s |
| Sustained decode | 251K | 4 | Valid | 289.4 | 72.5 | 2,202 ms | 13.8 ms | 4.0 | 0 | 8,680 | 30.0 s |
| Sustained decode | 251K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.