Speed test matrix
Ornith-1.5-9B-NVFP4 on RTX 3090 24GB
vllm · 0.32.0 · NVFP46 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K
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
- NVFP4
- 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: MTP, 2 tokens lookahead
# 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 ornith-ai/Ornith-1.5-9B-NVFP4 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}Prefill
prompt tok/s by context length8K context6,4351,273 ms TTFT
16K context5,9812,713 ms TTFT
32K context5,4145,965 ms TTFT
64K context4,57614,077 ms TTFT
128K context3,46037,191 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 | 156.252 ms TTFT | 249.059 ms TTFT | 469.2114 ms TTFT | 792.0160 ms TTFT | 1,372.9287 ms TTFT | 1,671.4459 ms TTFT | 1,533.1Requests queued843 ms TTFT | Over capacity |
| 16K | 121.6282 ms TTFT | 237.3492 ms TTFT | 486.7864 ms TTFT | 781.51,295 ms TTFT | 1,270.42,247 ms TTFT | Over capacity | Over capacity | Over capacity |
| 32K | 113.3344 ms TTFT | 244.1597 ms TTFT | 446.6984 ms TTFT | 724.91,545 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 64K | 117.5492 ms TTFT | 215.0826 ms TTFT | 371.01,336 ms TTFT | 592.12,029 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 112.0794 ms TTFT | 191.81,317 ms TTFT | 366.82,057 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 248K | 95.01,620 ms TTFT | 169.12,679 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
Requests queued: the server could not run all requests at once, so some waited in a queue. 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 | 6,435.0 | — | 1,273 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 5,981.0 | — | 2,713 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 5,414.0 | — | 5,965 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 4,576.0 | — | 14,077 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 3,460.0 | — | 37,191 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 156.2 | 155.7 | 52 ms | 6.4 ms | 1.0 | 0 | 4,684 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 249.0 | 129.0 | 59 ms | 7.8 ms | 2.0 | 0 | 7,467 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 469.2 | 120.7 | 114 ms | 8.3 ms | 4.0 | 0 | 14,076 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 792.0 | 105.7 | 160 ms | 9.5 ms | 8.0 | 0 | 23,750 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 1,372.9 | 87.9 | 287 ms | 11.4 ms | 16.0 | 0 | 41,190 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,671.4 | 53.4 | 459 ms | 18.7 ms | 32.0 | 0 | 50,128 | 30.0 s |
| Sustained decode | 0 | 64 | Requests queued | 1,533.1 | 32.2 | 843 ms | 31.1 ms | 52.5 | 0 | 45,945 | 30.0 s |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 121.6 | 124.9 | 282 ms | 8.0 ms | 1.0 | 0 | 3,648 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 237.3 | 121.7 | 492 ms | 8.2 ms | 2.0 | 0 | 7,119 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 486.7 | 120.7 | 864 ms | 8.3 ms | 4.0 | 0 | 14,595 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 781.5 | 100.7 | 1,295 ms | 9.9 ms | 8.0 | 0 | 23,445 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,270.4 | 76.8 | 2,247 ms | 13.0 ms | 16.0 | 0 | 38,112 | 30.0 s |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 113.3 | 117.5 | 344 ms | 8.5 ms | 1.0 | 0 | 3,398 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 244.1 | 124.3 | 597 ms | 8.0 ms | 2.0 | 0 | 7,322 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 446.6 | 113.2 | 984 ms | 8.8 ms | 4.0 | 0 | 13,397 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 724.9 | 91.8 | 1,545 ms | 10.9 ms | 8.0 | 0 | 21,747 | 30.0 s |
| 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 | 117.5 | 119.8 | 492 ms | 8.3 ms | 1.0 | 0 | 3,523 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 215.0 | 110.3 | 826 ms | 9.1 ms | 2.0 | 0 | 6,449 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 371.0 | 94.8 | 1,336 ms | 10.6 ms | 4.0 | 0 | 11,128 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 592.1 | 75.6 | 2,029 ms | 13.2 ms | 8.0 | 0 | 17,758 | 30.0 s |
| 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 | 128K | 1 | Valid | 112.0 | 114.6 | 794 ms | 8.7 ms | 1.0 | 0 | 3,360 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 191.8 | 99.1 | 1,317 ms | 10.1 ms | 2.0 | 0 | 5,753 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 366.8 | 91.2 | 2,057 ms | 11.0 ms | 4.0 | 0 | 10,998 | 30.0 s |
| Sustained decode | 128K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 1 | Valid | 95.0 | 96.7 | 1,620 ms | 10.3 ms | 1.0 | 0 | 2,849 | 30.0 s |
| Sustained decode | 248K | 2 | Valid | 169.1 | 86.4 | 2,679 ms | 11.6 ms | 2.0 | 0 | 5,074 | 30.0 s |
| Sustained decode | 248K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 248K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Requests queued: the server could not run all requests at once, so some waited in a queue. Over capacity: not run because this many requests at this context would exceed the server's KV cache.