Speed test matrix
NVIDIA-Nemotron-Nano-12B-v2-AWQ-4bit on RTX 3090 24GB
vllm · 0.32.0 · AWQ 4-bit5 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
- AWQ 4-bit
- 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. KV cache: int4_per_token_head # 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 cyankiwi/NVIDIA-Nemotron-Nano-12B-v2-AWQ-4bit --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN
Prefill
prompt tok/s by context length8K context4,0742,058 ms TTFT
16K context3,5274,710 ms TTFT
32K context2,83111,681 ms TTFT
64K context2,01932,670 ms TTFT
120K context1,34392,015 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 | 95.737 ms TTFT | 183.554 ms TTFT | 339.0113 ms TTFT | 590.8186 ms TTFT | 935.7335 ms TTFT | 1,343.2527 ms TTFT | 1,551.0675 ms TTFT | 1,272.1Requests queued732 ms TTFT |
| 16K | 88.3715 ms TTFT | 158.91,041 ms TTFT | 276.91,713 ms TTFT | 443.63,071 ms TTFT | 631.55,807 ms TTFT | 806.911,341 ms TTFT | Over capacity | Over capacity |
| 32K | 82.01,843 ms TTFT | 139.92,716 ms TTFT | 233.24,482 ms TTFT | 354.38,029 ms TTFT | 476.515,153 ms TTFT | Over capacity | Over capacity | Over capacity |
| 64K | 71.61,993 ms TTFT | 113.62,843 ms TTFT | 178.24,657 ms TTFT | 253.48,266 ms TTFT | 320.815,515 ms TTFT | Over capacity | Over capacity | Over capacity |
| 120K | 58.71,733 ms TTFT | 85.52,549 ms TTFT | 126.94,071 ms TTFT | 170.27,211 ms TTFT | 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 | 4,074.0 | — | 2,058 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 3,527.0 | — | 4,710 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 2,831.0 | — | 11,681 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 2,019.0 | — | 32,670 ms | — | — | 1 | — | — |
| Prefill | 120K | 1 | Valid | 1,343.0 | — | 92,015 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 95.7 | 95.8 | 37 ms | 10.4 ms | 1.0 | 0 | 2,872 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 183.5 | 91.9 | 54 ms | 10.9 ms | 2.0 | 0 | 5,504 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 339.0 | 85.0 | 113 ms | 11.8 ms | 4.0 | 0 | 10,167 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 590.8 | 74.2 | 186 ms | 13.5 ms | 8.0 | 0 | 17,720 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 935.7 | 58.8 | 335 ms | 17.0 ms | 16.0 | 0 | 28,064 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,343.2 | 42.1 | 527 ms | 23.8 ms | 32.0 | 0 | 40,288 | 30.0 s |
| Sustained decode | 0 | 64 | Valid | 1,551.0 | 23.9 | 675 ms | 41.8 ms | 64.0 | 0 | 46,531 | 30.0 s |
| Sustained decode | 0 | 128 | Requests queued | 1,272.1 | 19.3 | 732 ms | 51.8 ms | 67.0 | 0 | 38,173 | 30.0 s |
| Sustained decode | 16K | 1 | Valid | 88.3 | 88.4 | 715 ms | 11.3 ms | 1.0 | 0 | 2,649 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 158.9 | 78.8 | 1,041 ms | 12.7 ms | 2.0 | 0 | 4,768 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 276.9 | 67.4 | 1,713 ms | 14.8 ms | 4.0 | 0 | 8,304 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 443.6 | 52.0 | 3,071 ms | 19.2 ms | 8.0 | 0 | 13,304 | 30.0 s |
| Sustained decode | 16K | 16 | Valid | 631.5 | 34.7 | 5,807 ms | 28.8 ms | 16.0 | 0 | 18,944 | 30.0 s |
| Sustained decode | 16K | 32 | Valid | 806.9 | 19.7 | 11,341 ms | 50.8 ms | 32.0 | 0 | 24,192 | 30.0 s |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 82.0 | 82.0 | 1,843 ms | 12.2 ms | 1.0 | 0 | 2,458 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 139.9 | 68.3 | 2,716 ms | 14.6 ms | 2.0 | 0 | 4,196 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 233.2 | 54.2 | 4,482 ms | 18.5 ms | 4.0 | 0 | 6,996 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 354.3 | 37.6 | 8,029 ms | 26.6 ms | 8.0 | 0 | 10,624 | 30.0 s |
| Sustained decode | 32K | 16 | Valid | 476.5 | 21.6 | 15,153 ms | 46.3 ms | 16.0 | 0 | 14,296 | 30.0 s |
| Sustained decode | 32K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 71.6 | 71.7 | 1,993 ms | 13.9 ms | 1.0 | 0 | 2,148 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 113.6 | 55.5 | 2,843 ms | 18.0 ms | 2.0 | 0 | 3,410 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 178.2 | 41.4 | 4,657 ms | 24.2 ms | 4.0 | 0 | 5,344 | 30.0 s |
| Sustained decode | 64K | 8 | Valid | 253.4 | 26.8 | 8,266 ms | 37.3 ms | 8.0 | 0 | 7,600 | 30.0 s |
| Sustained decode | 64K | 16 | Valid | 320.8 | 14.6 | 15,515 ms | 68.5 ms | 16.0 | 0 | 9,626 | 30.0 s |
| Sustained decode | 64K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 1 | Valid | 58.7 | 58.8 | 1,733 ms | 17.0 ms | 1.0 | 0 | 1,760 | 30.0 s |
| Sustained decode | 120K | 2 | Valid | 85.5 | 41.9 | 2,549 ms | 23.9 ms | 2.0 | 0 | 2,565 | 30.0 s |
| Sustained decode | 120K | 4 | Valid | 126.9 | 29.8 | 4,071 ms | 33.5 ms | 4.0 | 0 | 3,808 | 30.0 s |
| Sustained decode | 120K | 8 | Valid | 170.2 | 18.4 | 7,211 ms | 54.2 ms | 8.0 | 0 | 5,104 | 30.0 s |
| Sustained decode | 120K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 120K | 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.