Speed test matrix
gemma-4-12B-it-qat-w4a16-ct on RTX 3090 24GB
vllm · 0.32.0 · W4A166 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
- AMD Ryzen 9 7900X (12C)
- RAM
- 32 GB
- OS
- Ubuntu 24.04.5
Engine
- Engine
- vllm
- Version
- 0.32.0
- Quantization
- W4A16
- 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), 350 W limit (VBIOS max), PCIe Gen4 x16 | AMD Ryzen 9 7900X 12C | 32 GB DDR5 | Ubuntu 24.04.5, NVIDIA driver 595.99.02 vllm serve google/gemma-4-12B-it-qat-w4a16-ct --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only
Prefill
prompt tok/s by context length8K context3,2762,446 ms TTFT
16K context2,8455,580 ms TTFT
32K context2,23214,157 ms TTFT
64K context1,59739,481 ms TTFT
128K context1,004125,423 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 | 80.537 ms TTFT | 154.730 ms TTFT | 286.237 ms TTFT | 500.155 ms TTFT | 796.6108 ms TTFT | 1,162.6143 ms TTFT | 1,499.1Requests queued221 ms TTFT | Over capacity |
| 16K | Looped | 133.286 ms TTFT | Looped | Looped | Looped | Over capacity | Over capacity | Over capacity |
| 32K | 61.5116 ms TTFT | Looped | Looped | Looped | Over capacity | Over capacity | Over capacity | Over capacity |
| 64K | 55.1215 ms TTFT | Looped | Looped | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 128K | 45.8438 ms TTFT | 90.3584 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 248K | 34.3Requests queued1,012 ms TTFT | 66.6Requests queued1,341 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. 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 | 3,276.0 | — | 2,446 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 2,845.0 | — | 5,580 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 2,232.0 | — | 14,157 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 1,597.0 | — | 39,481 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 1,004.0 | — | 125,423 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 80.5 | 81.0 | 37 ms | 12.4 ms | 1.0 | 0 | 2,414 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 154.7 | 78.1 | 30 ms | 12.8 ms | 2.0 | 0 | 4,640 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 286.2 | 72.7 | 37 ms | 13.8 ms | 4.0 | 0 | 8,584 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 500.1 | 64.3 | 55 ms | 15.6 ms | 8.0 | 0 | 15,000 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 796.6 | 52.0 | 108 ms | 19.2 ms | 16.0 | 0 | 23,888 | 30.0 s |
| Sustained decode | 0 | 32 | Valid | 1,162.6 | 39.2 | 143 ms | 25.5 ms | 32.0 | 0 | 34,848 | 30.0 s |
| Sustained decode | 0 | 64 | Requests queued | 1,499.1 | 30.3 | 221 ms | 33.0 ms | 52.1 | 0 | 44,961 | 30.0 s |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 2 | Valid | 133.2 | 66.9 | 86 ms | 15.0 ms | 2.0 | 0 | 3,994 | 30.0 s |
| Sustained decode | 16K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 16 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 61.5 | 61.5 | 116 ms | 16.3 ms | 1.0 | 0 | 1,844 | 30.0 s |
| Sustained decode | 32K | 2 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 32K | 8 | Looped | — | — | — | — | — | 0 | 0 | — |
| 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 | 55.1 | 55.2 | 215 ms | 18.1 ms | 1.0 | 0 | 1,652 | 30.0 s |
| Sustained decode | 64K | 2 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 64K | 4 | Looped | — | — | — | — | — | 0 | 0 | — |
| 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 | 128K | 1 | Valid | 45.8 | 45.9 | 438 ms | 21.8 ms | 1.0 | 0 | 1,372 | 30.0 s |
| Sustained decode | 128K | 2 | Valid | 90.3 | 45.2 | 584 ms | 22.1 ms | 2.0 | 0 | 2,708 | 30.0 s |
| Sustained decode | 128K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | Requests queued | 34.3 | 34.8 | 1,012 ms | 28.7 ms | 1.0 | 0 | 1,028 | 30.0 s |
| Sustained decode | 248K | 2 | Requests queued | 66.6 | 34.2 | 1,341 ms | 29.2 ms | 1.9 | 0 | 1,998 | 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. 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.