Speed test matrix
Qwen3.8-27B-GPTQ-4bit on RTX 3090 24GB
vllm · 0.26.0 · GPTQ3 contexts (0–16K) × 3 concurrency levels (1–4) · prefill at 4K, 8K, 16K
Hardware
- Type
- Discrete GPU
- GPU
- RTX 3090
- VRAM
- 24 GB
- CPU
- AMD Ryzen 9 9950X 16-Core Processor
- OS
- linux
Engine
- Engine
- vllm
- Version
- 0.26.0
- Quantization
- GPTQ
Run
- Served model
- btbtyler09/Qwen3.8-27B-GPTQ-4bit
- Revision
- main
- Benchmark tool
- llm-inference-bench v0.7.7
- Decode test
- Fixed duration
- Prefill mode
- Measured during decode warm-up
- Time per measurement
- 15 s
- Max output
- 8,192 tokens
Launch command
/home/lotto/vllm/bin/vllm serve /home/lotto/models/Qwen3.8-27B-GPTQ-4bit --served-model-name btbtyler09/Qwen3.8-27B-GPTQ-4bit --host 127.0.0.1 --port 42753 --max-model-len 24576 --gpu-memory-utilization 0.95 --max-num-seqs 8 --limit-mm-per-prompt '{"image":0,"video":0}'Prefill
prompt tok/s by context length4K context1,2923,247 ms TTFT
8K context1,2666,474 ms TTFT
16K context1,23713,071 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 |
|---|---|---|---|
| 0 | 43.4135 ms TTFT | 80.6222 ms TTFT | Over capacity |
| 4K | 42.63,238 ms TTFT | 78.35,955 ms TTFT | Over capacity |
| 16K | 40.613,092 ms TTFT | 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 | 4K | 1 | Valid | 1,292.0 | — | 3,247 ms | — | — | 1 | — | — |
| Prefill | 8K | 1 | Valid | 1,266.0 | — | 6,474 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 1,237.0 | — | 13,071 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 43.4 | 43.4 | 135 ms | 23.0 ms | 1.0 | 0 | 650 | 15.0 s |
| Sustained decode | 0 | 2 | Valid | 80.6 | 40.3 | 222 ms | 24.8 ms | 2.0 | 0 | 1,209 | 15.0 s |
| Sustained decode | 0 | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 4K | 1 | Valid | 42.6 | 42.5 | 3,238 ms | 23.5 ms | 1.0 | 0 | 638 | 15.0 s |
| Sustained decode | 4K | 2 | Valid | 78.3 | 38.1 | 5,955 ms | 26.2 ms | 2.0 | 0 | 1,174 | 15.0 s |
| Sustained decode | 4K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 40.6 | 40.6 | 13,092 ms | 24.6 ms | 1.0 | 0 | 609 | 15.0 s |
| Sustained decode | 16K | 2 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 4 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.