Speed test matrix
DeepSeek-V4-Flash on H200 NVL 141GB ×2
vllm · 0.25.0 · MXFP4+FP88 contexts (8K–1019K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 256K, 512K, 1019K, 1024K
Hardware
- Type
- Discrete GPU
- GPU
- H200 NVL
- GPU count
- 2
- VRAM
- 282 GB (2 × 141 GB)
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.25.0
- Quantization
- MXFP4+FP8
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
Run
- Served model
- deepseek
- 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-deepseek-vllm/bin/vllm" serve deepseek-ai/DeepSeek-V4-Flash --served-model-name deepseek --trust-remote-code --tensor-parallel-size 2 --reasoning-parser deepseek_v4 --enable-auto-tool-choice --tool-call-parser deepseek_v4 --max-model-len 1048576 --gpu-memory-utilization 0.846011 --host 0.0.0.0 --port 8000 --api-key <redacted> --kv-cache-dtype fp8 --block-size 256 --tokenizer-mode deepseek_v4
Prefill
prompt tok/s by context length8K context14K584 ms TTFT
16K context14.9K1,090 ms TTFT
32K context15K2,153 ms TTFT
64K context15K4,314 ms TTFT
128K context14.3K9,052 ms TTFT
256K context12.6K20,482 ms TTFT
512K context10.1K50,933 ms TTFT
1019K context7,159143,365 ms TTFT
1024K context7,106145,136 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 | 120.159 ms TTFT | 220.949 ms TTFT | 378.2105 ms TTFT | 602.2145 ms TTFT | 909.2210 ms TTFT | 1,345.3367 ms TTFT | 1,968.3499 ms TTFT | 2,811.9840 ms TTFT |
| 16K | 119.595 ms TTFT | 220.2176 ms TTFT | 375.1201 ms TTFT | 604.9315 ms TTFT | 899.5405 ms TTFT | 1,332.2640 ms TTFT | 1,898.91,099 ms TTFT | 2,777.21,519 ms TTFT |
| 32K | 119.0149 ms TTFT | 218.4260 ms TTFT | 377.4277 ms TTFT | 596.1358 ms TTFT | 891.7533 ms TTFT | 1,294.3853 ms TTFT | 1,927.11,488 ms TTFT | 2,752.52,559 ms TTFT |
| 64K | 118.4209 ms TTFT | 216.4312 ms TTFT | 370.1360 ms TTFT | 594.5549 ms TTFT | 888.3876 ms TTFT | 1,299.61,418 ms TTFT | 1,854.62,699 ms TTFT | 2,707.45,085 ms TTFT |
| 128K | 117.5306 ms TTFT | 214.4460 ms TTFT | 366.7606 ms TTFT | 578.9935 ms TTFT | 862.71,783 ms TTFT | 1,231.93,278 ms TTFT | 1,753.26,023 ms TTFT | 2,595.911,604 ms TTFT |
| 256K | 115.5602 ms TTFT | 210.2982 ms TTFT | 357.51,249 ms TTFT | 562.42,011 ms TTFT | 850.43,383 ms TTFT | 1,211.16,101 ms TTFT | 1,713.711,808 ms TTFT | 2,457.423,291 ms TTFT |
| 512K | 112.11,270 ms TTFT | 203.71,842 ms TTFT | 344.52,686 ms TTFT | 536.74,439 ms TTFT | 795.97,864 ms TTFT | 1,111.915,128 ms TTFT | 1,557.229,747 ms TTFT | 2,171.862,718 ms TTFT |
| 1019K | 108.02,580 ms TTFT | 196.73,817 ms TTFT | 328.35,460 ms TTFT | 507.99,278 ms TTFT | 734.616,539 ms TTFT | 1,007.532,264 ms TTFT | 1,305.959,263 ms TTFT | 1,465.1Requests queued104,729 ms TTFT |
Requests queued: the server could not run all requests at once, so some waited in a queue.
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 | 14K | — | 584 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 14.9K | — | 1,090 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 15K | — | 2,153 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 15K | — | 4,314 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 14.3K | — | 9,052 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 12.6K | — | 20,482 ms | — | — | 1 | — | — |
| Prefill | 512K | 1 | Valid | 10.1K | — | 50,933 ms | — | — | 1 | — | — |
| Prefill | 1019K | 1 | Valid | 7,159.0 | — | 143,365 ms | — | — | 1 | — | — |
| Prefill | 1024K | 1 | Valid | 7,106.0 | — | 145,136 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 120.1 | 120.1 | 59 ms | 8.3 ms | 1.0 | 0 | 3,603 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 220.9 | 110.5 | 49 ms | 9.0 ms | 2.0 | 0 | 6,628 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 378.2 | 94.8 | 105 ms | 10.5 ms | 4.0 | 0 | 11,345 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 602.2 | 75.8 | 145 ms | 13.2 ms | 8.0 | 0 | 18,064 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 909.2 | 57.4 | 210 ms | 17.4 ms | 16.0 | 0 | 27,264 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 1,345.3 | 42.7 | 367 ms | 23.4 ms | 32.0 | 0 | 40,353 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 1,968.3 | 31.3 | 499 ms | 31.9 ms | 64.0 | 0 | 59,013 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 2,811.9 | 22.4 | 840 ms | 44.7 ms | 128.0 | 0 | 84,359 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 119.5 | 119.5 | 95 ms | 8.4 ms | 1.0 | 0 | 3,584 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 220.2 | 110.1 | 176 ms | 9.1 ms | 2.0 | 0 | 6,607 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 375.1 | 94.1 | 201 ms | 10.6 ms | 4.0 | 0 | 11,252 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 604.9 | 76.0 | 315 ms | 13.2 ms | 8.0 | 0 | 18,144 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 899.5 | 56.5 | 405 ms | 17.7 ms | 16.0 | 0 | 26,977 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 1,332.2 | 42.1 | 640 ms | 23.8 ms | 32.0 | 0 | 39,968 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 1,898.9 | 30.1 | 1,099 ms | 33.2 ms | 64.0 | 0 | 56,959 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 2,777.2 | 21.7 | 1,519 ms | 46.1 ms | 128.0 | 0 | 83,336 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 119.0 | 119.1 | 149 ms | 8.4 ms | 1.0 | 0 | 3,571 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 218.4 | 109.3 | 260 ms | 9.2 ms | 2.0 | 0 | 6,554 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 377.4 | 94.7 | 277 ms | 10.6 ms | 4.0 | 0 | 11,328 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 596.1 | 74.8 | 358 ms | 13.4 ms | 8.0 | 0 | 17,880 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 891.7 | 56.1 | 533 ms | 17.8 ms | 16.0 | 0 | 26,739 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 1,294.3 | 40.7 | 853 ms | 24.6 ms | 32.0 | 0 | 38,819 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 1,927.1 | 30.2 | 1,488 ms | 33.1 ms | 64.0 | 0 | 57,792 | 30.0 s |
| Sustained decode | 32K | 128 | Valid | 2,752.5 | 21.5 | 2,559 ms | 46.6 ms | 128.0 | 0 | 82,569 | 30.0 s |
| Sustained decode | 64K | 1 | Valid | 118.4 | 118.6 | 209 ms | 8.4 ms | 1.0 | 0 | 3,555 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 216.4 | 108.3 | 312 ms | 9.2 ms | 2.0 | 0 | 6,490 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 370.1 | 92.6 | 360 ms | 10.8 ms | 4.0 | 0 | 11,100 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 594.5 | 74.7 | 549 ms | 13.4 ms | 8.0 | 0 | 17,832 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 888.3 | 55.9 | 876 ms | 17.9 ms | 16.0 | 0 | 26,639 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 1,299.6 | 40.9 | 1,418 ms | 24.5 ms | 32.0 | 0 | 38,977 | 30.0 s |
| Sustained decode | 64K | 64 | Valid | 1,854.6 | 29.2 | 2,699 ms | 34.3 ms | 64.0 | 0 | 55,618 | 30.0 s |
| Sustained decode | 64K | 128 | Valid | 2,707.4 | 21.2 | 5,085 ms | 47.1 ms | 128.0 | 0 | 81,155 | 30.0 s |
| Sustained decode | 128K | 1 | Valid | 117.5 | 117.7 | 306 ms | 8.5 ms | 1.0 | 0 | 3,526 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 214.4 | 107.5 | 460 ms | 9.3 ms | 2.0 | 0 | 6,432 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 366.7 | 91.9 | 606 ms | 10.9 ms | 4.0 | 0 | 11,000 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 578.9 | 72.7 | 935 ms | 13.8 ms | 8.0 | 0 | 17,368 | 30.0 s |
| Sustained decode | 128K | 16 | Valid | 862.7 | 54.3 | 1,783 ms | 18.4 ms | 16.0 | 0 | 25,872 | 30.0 s |
| Sustained decode | 128K | 32 | Valid | 1,231.9 | 39.0 | 3,278 ms | 25.7 ms | 32.0 | 0 | 36,960 | 30.0 s |
| Sustained decode | 128K | 64 | Valid | 1,753.2 | 28.0 | 6,023 ms | 35.7 ms | 64.0 | 0 | 52,606 | 30.0 s |
| Sustained decode | 128K | 128 | Valid | 2,595.9 | 20.8 | 11,604 ms | 48.2 ms | 128.0 | 0 | 77,827 | 30.0 s |
| Sustained decode | 256K | 1 | Valid | 115.5 | 115.8 | 602 ms | 8.6 ms | 1.0 | 0 | 3,465 | 30.0 s |
| Sustained decode | 256K | 2 | Valid | 210.2 | 105.8 | 982 ms | 9.5 ms | 2.0 | 0 | 6,311 | 30.0 s |
| Sustained decode | 256K | 4 | Valid | 357.5 | 89.6 | 1,249 ms | 11.2 ms | 4.0 | 0 | 10,724 | 30.0 s |
| Sustained decode | 256K | 8 | Valid | 562.4 | 70.6 | 2,011 ms | 14.2 ms | 8.0 | 0 | 16,873 | 30.0 s |
| Sustained decode | 256K | 16 | Valid | 850.4 | 53.3 | 3,383 ms | 18.8 ms | 16.0 | 0 | 25,504 | 30.0 s |
| Sustained decode | 256K | 32 | Valid | 1,211.1 | 38.0 | 6,101 ms | 26.3 ms | 32.0 | 0 | 36,321 | 30.0 s |
| Sustained decode | 256K | 64 | Valid | 1,713.7 | 27.1 | 11,808 ms | 36.9 ms | 64.0 | 0 | 51,393 | 30.0 s |
| Sustained decode | 256K | 128 | Valid | 2,457.4 | 19.7 | 23,291 ms | 50.9 ms | 128.0 | 0 | 73,724 | 30.0 s |
| Sustained decode | 512K | 1 | Valid | 112.1 | 112.6 | 1,270 ms | 8.9 ms | 1.0 | 0 | 3,364 | 30.0 s |
| Sustained decode | 512K | 2 | Valid | 203.7 | 102.2 | 1,842 ms | 9.8 ms | 2.0 | 0 | 6,114 | 30.0 s |
| Sustained decode | 512K | 4 | Valid | 344.5 | 86.6 | 2,686 ms | 11.5 ms | 4.0 | 0 | 10,332 | 30.0 s |
| Sustained decode | 512K | 8 | Valid | 536.7 | 67.5 | 4,439 ms | 14.8 ms | 8.0 | 0 | 16,096 | 30.0 s |
| Sustained decode | 512K | 16 | Valid | 795.9 | 50.0 | 7,864 ms | 20.0 ms | 16.0 | 0 | 23,872 | 30.0 s |
| Sustained decode | 512K | 32 | Valid | 1,111.9 | 35.0 | 15,128 ms | 28.6 ms | 32.0 | 0 | 33,340 | 30.0 s |
| Sustained decode | 512K | 64 | Valid | 1,557.2 | 25.1 | 29,747 ms | 39.8 ms | 64.0 | 0 | 46,720 | 30.0 s |
| Sustained decode | 512K | 128 | Valid | 2,171.8 | 17.9 | 62,718 ms | 55.8 ms | 128.0 | 0 | 65,156 | 30.0 s |
| Sustained decode | 1019K | 1 | Valid | 108.0 | 108.7 | 2,580 ms | 9.2 ms | 1.0 | 0 | 3,241 | 30.0 s |
| Sustained decode | 1019K | 2 | Valid | 196.7 | 98.9 | 3,817 ms | 10.1 ms | 2.0 | 0 | 5,900 | 30.0 s |
| Sustained decode | 1019K | 4 | Valid | 328.3 | 82.8 | 5,460 ms | 12.1 ms | 4.0 | 0 | 9,852 | 30.0 s |
| Sustained decode | 1019K | 8 | Valid | 507.9 | 63.9 | 9,278 ms | 15.6 ms | 8.0 | 0 | 15,235 | 30.0 s |
| Sustained decode | 1019K | 16 | Valid | 734.6 | 46.3 | 16,539 ms | 21.6 ms | 16.0 | 0 | 22,033 | 30.0 s |
| Sustained decode | 1019K | 32 | Valid | 1,007.5 | 32.2 | 32,264 ms | 31.0 ms | 32.0 | 0 | 30,211 | 30.0 s |
| Sustained decode | 1019K | 64 | Valid | 1,305.9 | 21.8 | 59,263 ms | 45.8 ms | 63.7 | 4 | 39,159 | 30.0 s |
| Sustained decode | 1019K | 128 | Requests queued | 1,465.1 | 17.6 | 104,729 ms | 56.9 ms | 91.4 | 12 | 43,887 | 30.0 s |
Requests queued: the server could not run all requests at once, so some waited in a queue.
