Speed test matrix
Laguna-XS.2-INT4 on RTX PRO 6000 Blackwell Server Edition 96GB
vllm · 0.30.1rc1.dev452+g73a583112 · INT4+INT86 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K
Hardware
- Type
- Discrete GPU
- GPU
- RTX PRO 6000 Blackwell Server Edition
- VRAM
- 96 GB
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.30.1rc1.dev452+g73a583112
- Quantization
- INT4+INT8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
- Commit
73a5831127a9
Run
- Served model
- poolside
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.7.6
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 30 s
- Max output
- 4,096 tokens
Launch command
env MAX_JOBS=2 FLASHINFER_NVCC_THREADS=1 TORCHINDUCTOR_COMPILE_THREADS=1 "$HOME/env_rtxpro6k-poolside-vllm-73a5831127a9/bin/vllm" serve poolside/Laguna-XS.2-INT4 --served-model-name poolside --trust-remote-code --tensor-parallel-size 1 --reasoning-parser poolside_v1 --enable-auto-tool-choice --tool-call-parser poolside_v1 --max-model-len 262144 --gpu-memory-utilization 0.81 --host 0.0.0.0 --port 8000 --api-key <redacted> --default-chat-template-kwargs '{"enable_thinking":true}' --attention-backend FLASHINFER --moe-backend marlinPrefill
prompt tok/s by context length8K context33.3K247 ms TTFT
16K context31.2K521 ms TTFT
32K context27K1,195 ms TTFT
64K context20.9K3,077 ms TTFT
128K context14.2K9,067 ms TTFT
251K context8,72328,822 ms TTFT
256K context8,59429,834 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 | 196.028 ms TTFT | 311.250 ms TTFT | 573.158 ms TTFT | 923.878 ms TTFT | 1,494.5118 ms TTFT | 2,228.3223 ms TTFT | Looped | 5,189.9667 ms TTFT |
| 16K | 191.242 ms TTFT | 304.358 ms TTFT | 564.752 ms TTFT | 879.8105 ms TTFT | 1,375.2172 ms TTFT | 2,004.4332 ms TTFT | 3,099.0597 ms TTFT | Over capacity |
| 32K | 182.367 ms TTFT | Looped | 534.387 ms TTFT | 814.2187 ms TTFT | Looped | 1,692.4639 ms TTFT | Over capacity | Over capacity |
| 64K | 167.5123 ms TTFT | 275.2134 ms TTFT | Looped | 707.2370 ms TTFT | 994.9680 ms TTFT | 1,270.81,297 ms TTFT | Over capacity | Over capacity |
| 128K | 143.7248 ms TTFT | 245.2270 ms TTFT | 425.0482 ms TTFT | 549.9859 ms TTFT | 712.11,558 ms TTFT | Over capacity | Over capacity | Over capacity |
| 251K | 113.7559 ms TTFT | 202.7766 ms TTFT | 334.01,217 ms TTFT | 386.71,954 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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 | 33.3K | — | 247 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 31.2K | — | 521 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 27K | — | 1,195 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 20.9K | — | 3,077 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 14.2K | — | 9,067 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 8,723.0 | — | 28,822 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 8,594.0 | — | 29,834 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 196.0 | 196.4 | 28 ms | 5.1 ms | 1.0 | 1 | 5,881 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 311.2 | 156.5 | 50 ms | 6.4 ms | 2.0 | 2 | 9,334 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 573.1 | 145.1 | 58 ms | 6.9 ms | 4.0 | 4 | 17,192 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 923.8 | 116.2 | 78 ms | 8.6 ms | 8.0 | 0 | 27,712 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 1,494.5 | 94.4 | 118 ms | 10.6 ms | 16.0 | 0 | 44,835 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 2,228.3 | 70.6 | 223 ms | 14.2 ms | 32.0 | 0 | 66,849 | 30.0 s |
| Sustained decode | 8K | 64 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 8K | 128 | Valid | 5,189.9 | 41.4 | 667 ms | 24.2 ms | 128.0 | 0 | 155,648 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 191.2 | 191.7 | 42 ms | 5.2 ms | 1.0 | 1 | 5,736 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 304.3 | 153.2 | 58 ms | 6.5 ms | 1.9 | 2 | 9,126 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 564.7 | 142.6 | 52 ms | 7.0 ms | 4.0 | 4 | 16,941 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 879.8 | 110.6 | 105 ms | 9.0 ms | 8.0 | 0 | 26,392 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,375.2 | 86.8 | 172 ms | 11.5 ms | 16.0 | 0 | 41,248 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 2,004.4 | 63.4 | 332 ms | 15.8 ms | 32.0 | 0 | 60,128 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 3,099.0 | 49.2 | 597 ms | 20.3 ms | 64.0 | 0 | 92,928 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 182.3 | 182.9 | 67 ms | 5.5 ms | 1.0 | 1 | 5,468 | 30.0 s |
| Sustained decode | 32K | 2 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 4 | Valid | 534.3 | 135.3 | 87 ms | 7.4 ms | 4.0 | 4 | 16,026 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 814.2 | 102.2 | 187 ms | 9.8 ms | 8.0 | 0 | 24,424 | 30.0 s |
| Sustained decode | 32K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 32 | Valid | 1,692.4 | 53.4 | 639 ms | 18.7 ms | 32.0 | 0 | 50,771 | 30.0 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 167.5 | 168.5 | 123 ms | 5.9 ms | 1.0 | 1 | 5,024 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 275.2 | 139.0 | 134 ms | 7.2 ms | 2.0 | 2 | 8,254 | 30.0 s |
| Sustained decode | 64K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 8 | Valid | 707.2 | 88.6 | 370 ms | 11.3 ms | 8.0 | 0 | 21,216 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 994.9 | 62.5 | 680 ms | 16.0 ms | 16.0 | 0 | 29,840 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 1,270.8 | 40.2 | 1,297 ms | 24.9 ms | 32.0 | 0 | 38,112 | 30.0 s |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 143.7 | 145.7 | 248 ms | 6.9 ms | 1.0 | 1 | 4,310 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 245.2 | 126.6 | 270 ms | 7.9 ms | 2.0 | 2 | 7,355 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 425.0 | 106.4 | 482 ms | 9.4 ms | 4.0 | 0 | 12,748 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 549.9 | 68.9 | 859 ms | 14.5 ms | 8.0 | 0 | 16,496 | 30.0 s |
| Sustained decode | 128K | 16 | Valid | 712.1 | 44.7 | 1,558 ms | 22.4 ms | 16.0 | 0 | 21,360 | 30.0 s |
| Sustained decode | 128K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 1 | Valid | 113.7 | 114.0 | 559 ms | 8.8 ms | 1.0 | 0 | 3,411 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 202.7 | 101.6 | 766 ms | 9.8 ms | 2.0 | 0 | 6,080 | 30.0 s |
| Sustained decode | 251K | 4 | Valid | 334.0 | 83.7 | 1,217 ms | 11.9 ms | 4.0 | 0 | 10,016 | 30.0 s |
| Sustained decode | 251K | 8 | Valid | 386.7 | 48.6 | 1,954 ms | 20.6 ms | 8.0 | 0 | 11,600 | 30.0 s |
| Sustained decode | 251K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Looped: the model kept repeating the same text, so the benchmark discarded this measurement. Over capacity: not run because this many requests at this context would exceed the server's KV cache.