Speed test matrix
NVIDIA-Nemotron-3-Nano-4B-FP8 on RTX 3090 24GB
vllm · 0.32.0 · FP86 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
- FP8
- 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 nvidia/NVIDIA-Nemotron-3-Nano-4B-FP8 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code
Prefill
prompt tok/s by context length8K context11.6K723 ms TTFT
16K context11.4K1,457 ms TTFT
32K context10.8K3,076 ms TTFT
64K context9,5716,895 ms TTFT
128K context7,83916,816 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 | 157.532 ms TTFT | 304.432 ms TTFT | 564.3278 ms TTFT | 957.694 ms TTFT | 1,554.7167 ms TTFT | 2,227.9253 ms TTFT | 2,926.1362 ms TTFT | 3,229.6780 ms TTFT |
| 16K | 154.3130 ms TTFT | 297.0209 ms TTFT | 547.7287 ms TTFT | 920.1495 ms TTFT | 1,411.7920 ms TTFT | 1,910.01,791 ms TTFT | 2,354.43,758 ms TTFT | Over capacity |
| 32K | 150.7260 ms TTFT | 289.1383 ms TTFT | 519.2566 ms TTFT | 802.51,042 ms TTFT | 1,190.21,879 ms TTFT | 1,579.93,651 ms TTFT | Over capacity | Over capacity |
| 64K | 143.9365 ms TTFT | 276.4473 ms TTFT | 478.2741 ms TTFT | 701.11,202 ms TTFT | 975.92,126 ms TTFT | Over capacity | Over capacity | Over capacity |
| 128K | 131.4492 ms TTFT | 250.4568 ms TTFT | 410.1760 ms TTFT | 547.11,173 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 248K | 113.8951 ms TTFT | 217.01,152 ms TTFT | 323.51,572 ms TTFT | Over capacity | Over capacity | Over capacity | 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 | 11.6K | — | 723 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 11.4K | — | 1,457 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 10.8K | — | 3,076 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 9,571.0 | — | 6,895 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 7,839.0 | — | 16,816 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 157.5 | 157.9 | 32 ms | 6.3 ms | 1.0 | 0 | 4,724 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 304.4 | 152.3 | 32 ms | 6.6 ms | 2.0 | 0 | 9,130 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 564.3 | 141.2 | 278 ms | 7.1 ms | 4.0 | 0 | 16,927 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 957.6 | 120.0 | 94 ms | 8.3 ms | 8.0 | 0 | 28,728 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 1,554.7 | 97.3 | 167 ms | 10.3 ms | 16.0 | 0 | 46,640 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 2,227.9 | 69.9 | 253 ms | 14.3 ms | 32.0 | 0 | 66,816 | 30.0 s |
| Sustained decode | 0 | 64 | Valid | 2,926.1 | 45.7 | 362 ms | 21.9 ms | 64.0 | 0 | 87,744 | 30.0 s |
| Sustained decode | 0 | 128 | Valid | 3,229.6 | 25.1 | 780 ms | 39.9 ms | 128.0 | 0 | 96,896 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 154.3 | 154.5 | 130 ms | 6.5 ms | 1.0 | 0 | 4,629 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 297.0 | 148.5 | 209 ms | 6.7 ms | 2.0 | 0 | 8,910 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 547.7 | 136.5 | 287 ms | 7.3 ms | 4.0 | 0 | 16,428 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 920.1 | 114.2 | 495 ms | 8.8 ms | 8.0 | 0 | 27,600 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 1,411.7 | 86.8 | 920 ms | 11.5 ms | 16.0 | 0 | 42,353 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 1,910.0 | 57.6 | 1,791 ms | 17.4 ms | 32.0 | 0 | 57,280 | 30.0 s |
| Sustained decode | 16K | 64 | Valid | 2,354.4 | 34.3 | 3,758 ms | 29.1 ms | 64.0 | 0 | 70,592 | 30.0 s |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 150.7 | 150.8 | 260 ms | 6.6 ms | 1.0 | 0 | 4,520 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 289.1 | 144.3 | 383 ms | 6.9 ms | 2.0 | 0 | 8,672 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 519.2 | 128.7 | 566 ms | 7.8 ms | 4.0 | 0 | 15,576 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 802.5 | 98.6 | 1,042 ms | 10.1 ms | 8.0 | 0 | 24,072 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 1,190.2 | 71.4 | 1,879 ms | 14.0 ms | 16.0 | 0 | 35,696 | 30.0 s |
| Sustained decode | 32K | 32 | Valid | 1,579.9 | 45.5 | 3,651 ms | 22.0 ms | 32.0 | 0 | 47,392 | 30.0 s |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 143.9 | 144.0 | 365 ms | 6.9 ms | 1.0 | 0 | 4,317 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 276.4 | 138.0 | 473 ms | 7.2 ms | 2.0 | 0 | 8,291 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 478.2 | 118.6 | 741 ms | 8.4 ms | 4.0 | 0 | 14,344 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 701.1 | 85.9 | 1,202 ms | 11.6 ms | 8.0 | 0 | 21,032 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 975.9 | 58.2 | 2,126 ms | 17.2 ms | 16.0 | 0 | 29,280 | 30.0 s |
| Sustained decode | 64K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 131.4 | 131.8 | 492 ms | 7.6 ms | 1.0 | 0 | 3,940 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 250.4 | 125.0 | 568 ms | 8.0 ms | 2.0 | 0 | 7,510 | 30.0 s |
| Sustained decode | 128K | 4 | Valid | 410.1 | 102.0 | 760 ms | 9.8 ms | 4.0 | 0 | 12,300 | 30.0 s |
| Sustained decode | 128K | 8 | Valid | 547.1 | 67.4 | 1,173 ms | 14.8 ms | 8.0 | 0 | 16,408 | 30.0 s |
| 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 | 113.8 | 114.3 | 951 ms | 8.8 ms | 1.0 | 0 | 3,415 | 30.0 s |
| Sustained decode | 248K | 2 | Valid | 217.0 | 108.2 | 1,152 ms | 9.2 ms | 2.0 | 0 | 6,508 | 30.0 s |
| Sustained decode | 248K | 4 | Valid | 323.5 | 79.8 | 1,572 ms | 12.5 ms | 4.0 | 0 | 9,704 | 30.0 s |
| 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 | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.