Speed test matrix
Shieldstral-1.0-3B on H200 NVL 141GB
vllm · 0.26.1rc1.dev638+g52be12cfa · BF163 contexts (8K–27K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 27K, 32K
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
- shieldstral
- 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-mistralai-vllm/bin/vllm" serve mistralai/Shieldstral-1.0-3B --served-model-name shieldstral --tensor-parallel-size 1 --max-model-len 32768 --gpu-memory-utilization 0.874627 --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context61.6K133 ms TTFT
16K context50.6K322 ms TTFT
27K context40.7K671 ms TTFT
32K context39K831 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 | 289.835 ms TTFT | 554.454 ms TTFT | 1,031.662 ms TTFT | 1,752.792 ms TTFT | 2,623.4146 ms TTFT | 3,412.1266 ms TTFT | 4,360.4445 ms TTFT | Over capacity |
| 16K | 271.654 ms TTFT | 510.656 ms TTFT | 909.486 ms TTFT | 1,431.0128 ms TTFT | 1,923.4204 ms TTFT | 2,419.2217 ms TTFT | Over capacity | Over capacity |
| 27K | 250.661 ms TTFT | 460.771 ms TTFT | 778.9122 ms TTFT | 1,134.2151 ms TTFT | 1,444.7331 ms TTFT | 1,691.6587 ms TTFT | 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 | 61.6K | — | 133 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 50.6K | — | 322 ms | — | — | 1 | — | — |
| Prefill | 27K | 1 | Valid | 40.7K | — | 671 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 39K | — | 831 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 289.8 | 290.3 | 35 ms | 3.4 ms | 1.0 | 2 | 8,692 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 554.4 | 278.3 | 54 ms | 3.6 ms | 2.0 | 4 | 16,632 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 1,031.6 | 259.5 | 62 ms | 3.9 ms | 4.0 | 8 | 30,947 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 1,752.7 | 221.5 | 92 ms | 4.5 ms | 8.0 | 8 | 52,581 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 2,623.4 | 170.1 | 146 ms | 5.9 ms | 16.0 | 16 | 78,704 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 3,412.1 | 109.6 | 266 ms | 9.1 ms | 32.0 | 0 | 102,368 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 4,360.4 | 69.8 | 445 ms | 14.3 ms | 64.0 | 0 | 130,817 | 30.0 s |
| Sustained decode | 8K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 271.6 | 272.7 | 54 ms | 3.7 ms | 1.0 | 2 | 8,148 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 510.6 | 256.7 | 56 ms | 3.9 ms | 2.0 | 4 | 15,318 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 909.4 | 228.7 | 86 ms | 4.4 ms | 4.0 | 4 | 27,296 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 1,431.0 | 181.0 | 128 ms | 5.5 ms | 8.0 | 8 | 42,926 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,923.4 | 125.9 | 204 ms | 7.9 ms | 16.0 | 16 | 57,699 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 2,419.2 | 76.7 | 217 ms | 13.0 ms | 32.0 | 0 | 72,576 | 30.0 s |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 27K | 1 | Valid | 250.6 | 251.9 | 61 ms | 4.0 ms | 1.0 | 2 | 7,517 | 30.0 s |
| Sustained decode | 27K | 2 | Valid | 460.7 | 231.7 | 71 ms | 4.3 ms | 2.0 | 2 | 13,820 | 30.0 s |
| Sustained decode | 27K | 4 | Valid | 778.9 | 196.3 | 122 ms | 5.1 ms | 4.0 | 4 | 23,366 | 30.0 s |
| Sustained decode | 27K | 8 | Valid | 1,134.2 | 143.5 | 151 ms | 7.0 ms | 7.7 | 8 | 34,026 | 30.0 s |
| Sustained decode | 27K | 16 | Valid | 1,444.7 | 91.3 | 331 ms | 11.0 ms | 16.0 | 0 | 43,344 | 30.0 s |
| Sustained decode | 27K | 32 | Valid | 1,691.6 | 53.3 | 587 ms | 18.8 ms | 32.0 | 0 | 50,752 | 30.0 s |
| Sustained decode | 27K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 27K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.