Speed test matrix
Ling-3.0-tiny-base-midtrain on H200 NVL 141GB
sglang · 0.5.6.post3.dev8719+ge1a24a189 · 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
- sglang
- Version
- 0.5.6.post3.dev8719+ge1a24a189
- Quantization
- BF16
- Backend
- cuda
- Repository
- github.com/sgl-project/sglang
- Build
- cuda-13.2
- Commit
e1a24a189fe4
Run
- Served model
- ling
- 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-inclusionai-sglang/bin/sglang" serve --model-path inclusionAI/Ling-3.0-tiny-base-midtrain --served-model-name ling --trust-remote-code --tp 1 --reasoning-parser ling3 --tool-call-parser ling3 --context-length 262144 --mem-fraction-static 0.869695 --enable-metrics --host 0.0.0.0 --port 8000 --api-key <redacted>
Prefill
prompt tok/s by context length8K context64K128 ms TTFT
16K context38K426 ms TTFT
32K context78.4K411 ms TTFT
64K context71K907 ms TTFT
128K context59K2,180 ms TTFT
251K context42.2K5,978 ms TTFT
256K context41.5K6,200 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 | 403.663 ms TTFT | 754.874 ms TTFT | 1,240.388 ms TTFT | 1,983.5143 ms TTFT | 2,989.4220 ms TTFT | 5,062.8392 ms TTFT | 7,260.1759 ms TTFT | 12.6K1,332 ms TTFT |
| 16K | 399.177 ms TTFT | 712.1110 ms TTFT | 1,257.499 ms TTFT | 1,923.4232 ms TTFT | 2,984.9405 ms TTFT | 4,546.3733 ms TTFT | 5,973.01,348 ms TTFT | 10.4K2,652 ms TTFT |
| 32K | 391.7108 ms TTFT | 519.4150 ms TTFT | 1,215.6150 ms TTFT | 1,892.4380 ms TTFT | 2,799.6551 ms TTFT | 3,888.31,478 ms TTFT | 6,491.92,985 ms TTFT | 8,981.25,715 ms TTFT |
| 64K | 379.8159 ms TTFT | 326.3257 ms TTFT | 1,099.4263 ms TTFT | 1,621.4681 ms TTFT | 2,478.01,285 ms TTFT | 2,391.73,221 ms TTFT | 4,324.16,372 ms TTFT | Over capacity |
| 128K | 362.3266 ms TTFT | 238.5459 ms TTFT | 1,017.0437 ms TTFT | 1,342.71,120 ms TTFT | 1,874.23,207 ms TTFT | 1,878.76,348 ms TTFT | 2,161.812,624 ms TTFT | Over capacity |
| 251K | 329.5510 ms TTFT | 69.8891 ms TTFT | 141.71,638 ms TTFT | 980.83,247 ms TTFT | 1,275.96,161 ms TTFT | 1,602.112,879 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 | 64K | — | 128 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 38K | — | 426 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 78.4K | — | 411 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 71K | — | 907 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 59K | — | 2,180 ms | — | — | 1 | — | — |
| Prefill | 251K | 1 | Valid | 42.2K | — | 5,978 ms | — | — | 1 | — | — |
| Prefill | 256K | 1 | Valid | 41.5K | — | 6,200 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 403.6 | 405.9 | 63 ms | 2.5 ms | 1.0 | 3 | 12,107 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 754.8 | 380.1 | 74 ms | 2.6 ms | 2.0 | 6 | 22,644 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 1,240.3 | 322.3 | 88 ms | 3.1 ms | 4.3 | 8 | 37,210 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 1,983.5 | 252.3 | 143 ms | 4.0 ms | 8.1 | 16 | 59,504 | 30.0 s |
| Sustained decode | 8K | 16 | Valid | 2,989.4 | 189.9 | 220 ms | 5.3 ms | 19.7 | 16 | 89,680 | 30.0 s |
| Sustained decode | 8K | 32 | Valid | 5,062.8 | 159.3 | 392 ms | 6.3 ms | 34.1 | 32 | 151,888 | 30.0 s |
| Sustained decode | 8K | 64 | Valid | 7,260.1 | 113.0 | 759 ms | 8.8 ms | 84.3 | 0 | 217,808 | 30.0 s |
| Sustained decode | 8K | 128 | Valid | 12.6K | 98.4 | 1,332 ms | 10.2 ms | 128.0 | 0 | 377,854 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 399.1 | 402.0 | 77 ms | 2.5 ms | 1.0 | 3 | 11,974 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 712.1 | 375.0 | 110 ms | 2.7 ms | 10.8 | 4 | 21,362 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 1,257.4 | 317.1 | 99 ms | 3.2 ms | 4.0 | 8 | 37,749 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 1,923.4 | 239.2 | 232 ms | 4.2 ms | 7.7 | 8 | 57,697 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 2,984.9 | 189.7 | 405 ms | 5.3 ms | 15.5 | 16 | 89,539 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 4,546.3 | 150.3 | 733 ms | 6.7 ms | 35.8 | 32 | 136,386 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 5,973.0 | 93.6 | 1,348 ms | 10.7 ms | 94.9 | 0 | 179,199 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 10.4K | 79.2 | 2,652 ms | 12.6 ms | 142.9 | 0 | 312,832 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 391.7 | 396.3 | 108 ms | 2.5 ms | 1.0 | 3 | 11,752 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 519.4 | 271.7 | 150 ms | 4.2 ms | 48.9 | 2 | 15,582 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 1,215.6 | 306.9 | 150 ms | 3.3 ms | 3.9 | 8 | 36,469 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 1,892.4 | 234.7 | 380 ms | 4.3 ms | 8.1 | 8 | 56,773 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 2,799.6 | 177.2 | 551 ms | 5.6 ms | 15.9 | 16 | 83,985 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 3,888.3 | 120.3 | 1,478 ms | 8.3 ms | 39.4 | 0 | 116,639 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 6,491.9 | 104.5 | 2,985 ms | 9.6 ms | 64.0 | 0 | 194,767 | 30.0 s |
| Sustained decode | 32K | 128 | Valid | 8,981.2 | 69.5 | 5,715 ms | 14.4 ms | 128.6 | 0 | 269,439 | 30.0 s |
| Sustained decode | 64K | 1 | Valid | 379.8 | 386.2 | 159 ms | 2.6 ms | 0.9 | 3 | 11,392 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 326.3 | 243.6 | 257 ms | 5.2 ms | 84.5 | 2 | 9,788 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 1,099.4 | 287.5 | 263 ms | 3.5 ms | 4.3 | 8 | 32,982 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 1,621.4 | 207.4 | 681 ms | 4.8 ms | 9.1 | 8 | 48,630 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 2,478.0 | 159.7 | 1,285 ms | 6.3 ms | 16.1 | 16 | 74,337 | 30.0 s |
| Sustained decode | 64K | 32 | Valid | 2,391.7 | 77.8 | 3,221 ms | 12.8 ms | 46.9 | 0 | 71,742 | 30.0 s |
| Sustained decode | 64K | 64 | Valid | 4,324.1 | 70.3 | 6,372 ms | 14.2 ms | 77.3 | 0 | 129,728 | 30.0 s |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 362.3 | 368.4 | 266 ms | 2.7 ms | 1.0 | 2 | 10,870 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 238.5 | 111.7 | 459 ms | 9.0 ms | 49.5 | 0 | 7,156 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 1,017.0 | 262.1 | 437 ms | 3.8 ms | 3.9 | 8 | 30,510 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 1,342.7 | 178.8 | 1,120 ms | 5.6 ms | 9.8 | 8 | 40,278 | 30.0 s |
| Sustained decode | 128K | 16 | Valid | 1,874.2 | 120.4 | 3,207 ms | 8.3 ms | 19.4 | 5 | 56,077 | 29.9 s |
| Sustained decode | 128K | 32 | Valid | 1,878.7 | 65.5 | 6,348 ms | 15.3 ms | 37.0 | 0 | 56,352 | 30.0 s |
| Sustained decode | 128K | 64 | Valid | 2,161.8 | 39.0 | 12,624 ms | 25.6 ms | 96.0 | 0 | 64,832 | 30.0 s |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 251K | 1 | Valid | 329.5 | 341.1 | 510 ms | 2.9 ms | 0.9 | 2 | 9,884 | 30.0 s |
| Sustained decode | 251K | 2 | Valid | 69.8 | 34.9 | 891 ms | 28.6 ms | 66.4 | 0 | 2,094 | 30.0 s |
| Sustained decode | 251K | 4 | Valid | 141.7 | 35.8 | 1,638 ms | 28.0 ms | 69.0 | 0 | 4,248 | 30.0 s |
| Sustained decode | 251K | 8 | Valid | 980.8 | 126.7 | 3,247 ms | 7.9 ms | 10.8 | 6 | 29,001 | 29.6 s |
| Sustained decode | 251K | 16 | Valid | 1,275.9 | 85.7 | 6,161 ms | 11.7 ms | 22.0 | 0 | 38,272 | 30.0 s |
| Sustained decode | 251K | 32 | Valid | 1,602.1 | 54.4 | 12,879 ms | 18.4 ms | 36.8 | 0 | 48,064 | 30.0 s |
| 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.
