Speed test matrix
gpt-oss-20b on RTX 3090 24GB
vllm · 0.32.0 · MXFP45 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K
Hardware
- Type
- Discrete GPU
- GPU
- RTX 3090
- VRAM
- 24 GB
- CPU
- Intel Core i7-9700K (8C)
- RAM
- 32 GB
- OS
- Ubuntu 24.04.5
Engine
- Engine
- vllm
- Version
- 0.32.0
- Quantization
- MXFP4
- Backend
- cuda
- Repository
- github.com/vllm-project/vllm
- Build
- vllm/vllm-openai:v0.32.0-x86_64
Run
- Served model
- bench
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.8.0
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 30 s
- Max output
- 8,192 tokens
Launch command
# Untuned out-of-the-box baseline. Spec decode: none # 1x RTX 3090 24GB (Zagitta/gpu-unlock), 420 W limit, PCIe Gen3 x16 | Intel Core i7-9700K 8C | 32 GB DDR4 | Ubuntu 24.04.5, NVIDIA driver 580.178.04 vllm serve openai/gpt-oss-20b --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code
Prefill
prompt tok/s by context length8K context10.5K775 ms TTFT
16K context8,4251,911 ms TTFT
32K context6,0655,275 ms TTFT
64K context3,87216,471 ms TTFT
120K context2,36950,393 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 | 32 | 64 | 128 |
|---|---|---|---|---|---|---|---|---|
| 0 | Looped | 268.837 ms TTFT | 436.253 ms TTFT | 565.364 ms TTFT | 879.993 ms TTFT | 1,813.0116 ms TTFT | Over capacity | Over capacity |
| 16K | 150.168 ms TTFT | 230.785 ms TTFT | 351.0131 ms TTFT | 534.8240 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 32K | 136.1100 ms TTFT | 215.9140 ms TTFT | 293.1232 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 64K | 111.1168 ms TTFT | 174.5250 ms TTFT | 217.6368 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 120K | 85.3284 ms TTFT | 135.2492 ms TTFT | Over capacity | Over capacity | 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 | 10.5K | — | 775 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 8,425.0 | — | 1,911 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 6,065.0 | — | 5,275 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 3,872.0 | — | 16,471 ms | — | — | 1 | — | — |
| Prefill | 120K | 1 | Valid | 2,369.0 | — | 50,393 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 0 | 2 | Valid | 268.8 | 131.1 | 37 ms | 7.6 ms | 2.0 | 0 | 8,065 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 436.2 | 107.2 | 53 ms | 9.3 ms | 4.0 | 0 | 13,090 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 565.3 | 83.9 | 64 ms | 11.9 ms | 8.0 | 0 | 16,956 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 879.9 | 68.9 | 93 ms | 14.5 ms | 16.0 | 0 | 26,396 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,813.0 | 61.7 | 116 ms | 16.2 ms | 32.0 | 0 | 54,381 | 30.0 s |
| Sustained decode | 0 | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 150.1 | 150.5 | 68 ms | 6.6 ms | 1.0 | 0 | 4,501 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 230.7 | 116.2 | 85 ms | 8.6 ms | 2.0 | 0 | 6,922 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 351.0 | 86.1 | 131 ms | 11.6 ms | 4.0 | 0 | 10,528 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 534.8 | 67.9 | 240 ms | 14.7 ms | 8.0 | 0 | 16,044 | 30.0 s |
| Sustained decode | 16K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 136.1 | 134.6 | 100 ms | 7.4 ms | 1.0 | 0 | 4,084 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 215.9 | 104.9 | 140 ms | 9.5 ms | 2.0 | 0 | 6,478 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 293.1 | 76.9 | 232 ms | 13.0 ms | 4.0 | 0 | 8,792 | 30.0 s |
| Sustained decode | 32K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 111.1 | 111.4 | 168 ms | 9.0 ms | 1.0 | 0 | 3,333 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 174.5 | 87.8 | 250 ms | 11.4 ms | 2.0 | 0 | 5,233 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 217.6 | 59.8 | 368 ms | 16.7 ms | 4.0 | 0 | 6,240 | 28.7 s |
| Sustained decode | 64K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 1 | Valid | 85.3 | 85.6 | 284 ms | 11.7 ms | 1.0 | 0 | 2,560 | 30.0 s |
| Sustained decode | 120K | 2 | Valid | 135.2 | 68.2 | 492 ms | 14.7 ms | 2.0 | 0 | 4,056 | 30.0 s |
| Sustained decode | 120K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 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.