Speed test matrix
Ling-3.0-tiny-int4 on RTX PRO 6000 Blackwell Server Edition 96GB
vllm · 0.30.1rc1.dev452+g73a583112 · INT45 contexts (8K–123K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 123K, 128K
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
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- cuda-13.2
- Commit
73a5831127a9
Run
- Served model
- ling
- 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 "$HOME/env_rtxpro6k-inclusionai-vllm-73a5831127a9/bin/vllm" serve inclusionAI/Ling-3.0-tiny-int4 --served-model-name ling --trust-remote-code --tensor-parallel-size 1 --reasoning-parser ling3 --enable-auto-tool-choice --tool-call-parser ling3 --max-model-len 131072 --gpu-memory-utilization 0.75 --host 0.0.0.0 --port 8000 --api-key <redacted> --mamba-cache-mode align --attention-backend TRITON_MLA
Prefill
prompt tok/s by context length8K context55K149 ms TTFT
16K context57.5K282 ms TTFT
32K context55K587 ms TTFT
64K context46.4K1,388 ms TTFT
123K context35.4K3,490 ms TTFT
128K context34.6K3,722 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 | 325.241 ms TTFT | 538.869 ms TTFT | 963.970 ms TTFT | 1,627.8112 ms TTFT | Looped | 3,520.8248 ms TTFT | Looped | 6,091.1783 ms TTFT |
| 16K | Looped | 533.1117 ms TTFT | 938.3150 ms TTFT | Looped | 2,199.0364 ms TTFT | Looped | 4,207.1934 ms TTFT | 5,319.41,644 ms TTFT |
| 32K | 311.5101 ms TTFT | 516.1159 ms TTFT | 892.0218 ms TTFT | 1,401.8314 ms TTFT | Looped | 2,828.9774 ms TTFT | 3,541.51,319 ms TTFT | 4,435.82,442 ms TTFT |
| 64K | 294.4165 ms TTFT | 484.9241 ms TTFT | 810.2333 ms TTFT | 1,222.2478 ms TTFT | Looped | 2,250.31,421 ms TTFT | 2,744.92,672 ms TTFT | 3,314.25,169 ms TTFT |
| 123K | 265.1306 ms TTFT | 441.2353 ms TTFT | 685.8484 ms TTFT | 988.7358 ms TTFT | 1,330.01,574 ms TTFT | 1,699.33,126 ms TTFT | 1,967.75,976 ms TTFT | 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 | 55K | — | 149 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 57.5K | — | 282 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 55K | — | 587 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 46.4K | — | 1,388 ms | — | — | 1 | — | — |
| Prefill | 123K | 1 | Valid | 35.4K | — | 3,490 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 34.6K | — | 3,722 ms | — | — | 1 | — | — |
| Sustained decode | 8K | 1 | Valid | 325.2 | 326.1 | 41 ms | 3.1 ms | 1.0 | 2 | 9,756 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 538.8 | 270.8 | 69 ms | 3.7 ms | 2.0 | 4 | 16,164 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 963.9 | 242.3 | 70 ms | 4.1 ms | 4.0 | 8 | 28,914 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 1,627.8 | 204.8 | 112 ms | 4.9 ms | 8.0 | 8 | 48,831 | 30.0 s |
| Sustained decode | 8K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 8K | 32 | Valid | 3,520.8 | 110.5 | 248 ms | 9.0 ms | 32.0 | 0 | 105,604 | 30.0 s |
| Sustained decode | 8K | 64 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 8K | 128 | Valid | 6,091.1 | 47.4 | 783 ms | 21.1 ms | 128.0 | 0 | 182,670 | 30.0 s |
| Sustained decode | 16K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 2 | Valid | 533.1 | 268.6 | 117 ms | 3.7 ms | 1.9 | 4 | 15,992 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 938.3 | 236.1 | 150 ms | 4.2 ms | 4.0 | 4 | 28,148 | 30.0 s |
| Sustained decode | 16K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 16 | Valid | 2,199.0 | 139.6 | 364 ms | 7.2 ms | 15.6 | 16 | 65,976 | 30.0 s |
| Sustained decode | 16K | 32 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 64 | Valid | 4,207.1 | 65.7 | 934 ms | 15.2 ms | 64.0 | 0 | 126,208 | 30.0 s |
| Sustained decode | 16K | 128 | Valid | 5,319.4 | 41.2 | 1,644 ms | 24.3 ms | 128.0 | 0 | 159,629 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 311.5 | 314.9 | 101 ms | 3.2 ms | 1.0 | 2 | 9,343 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 516.1 | 260.9 | 159 ms | 3.8 ms | 2.0 | 4 | 15,487 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 892.0 | 225.0 | 218 ms | 4.4 ms | 4.0 | 4 | 26,760 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 1,401.8 | 178.1 | 314 ms | 5.6 ms | 7.8 | 8 | 42,048 | 30.0 s |
| Sustained decode | 32K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 32 | Valid | 2,828.9 | 88.4 | 774 ms | 11.3 ms | 32.0 | 0 | 84,868 | 30.0 s |
| Sustained decode | 32K | 64 | Valid | 3,541.5 | 54.7 | 1,319 ms | 18.3 ms | 64.0 | 0 | 106,243 | 30.0 s |
| Sustained decode | 32K | 128 | Valid | 4,435.8 | 33.7 | 2,442 ms | 29.7 ms | 128.0 | 0 | 132,995 | 30.0 s |
| Sustained decode | 64K | 1 | Valid | 294.4 | 297.7 | 165 ms | 3.4 ms | 1.0 | 2 | 8,832 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 484.9 | 246.6 | 241 ms | 4.1 ms | 2.0 | 4 | 14,549 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 810.2 | 204.8 | 333 ms | 4.9 ms | 4.0 | 4 | 24,305 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 1,222.2 | 155.3 | 478 ms | 6.4 ms | 8.0 | 8 | 36,659 | 30.0 s |
| Sustained decode | 64K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 32 | Valid | 2,250.3 | 69.0 | 1,421 ms | 14.5 ms | 32.0 | 0 | 67,491 | 30.0 s |
| Sustained decode | 64K | 64 | Valid | 2,744.9 | 41.1 | 2,672 ms | 24.4 ms | 64.0 | 0 | 82,306 | 30.0 s |
| Sustained decode | 64K | 128 | Valid | 3,314.2 | 24.0 | 5,169 ms | 41.6 ms | 128.0 | 0 | 99,427 | 30.0 s |
| Sustained decode | 123K | 1 | Valid | 265.1 | 270.6 | 306 ms | 3.7 ms | 1.0 | 2 | 7,954 | 30.0 s |
| Sustained decode | 123K | 2 | Valid | 441.2 | 223.3 | 353 ms | 4.5 ms | 2.0 | 2 | 13,233 | 30.0 s |
| Sustained decode | 123K | 4 | Valid | 685.8 | 173.8 | 484 ms | 5.8 ms | 4.0 | 4 | 20,575 | 30.0 s |
| Sustained decode | 123K | 8 | Valid | 988.7 | 125.3 | 358 ms | 8.0 ms | 7.9 | 8 | 29,662 | 30.0 s |
| Sustained decode | 123K | 16 | Valid | 1,330.0 | 82.0 | 1,574 ms | 12.2 ms | 16.0 | 0 | 39,888 | 30.0 s |
| Sustained decode | 123K | 32 | Valid | 1,699.3 | 51.9 | 3,126 ms | 19.3 ms | 32.0 | 0 | 50,978 | 30.0 s |
| Sustained decode | 123K | 64 | Valid | 1,967.7 | 29.6 | 5,976 ms | 33.8 ms | 64.0 | 0 | 59,012 | 30.0 s |
| Sustained decode | 123K | 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.