Speed test matrix
Fara1.5-27B-AWQ-W4A16 on RTX 3090 24GB
vllm · 0.32.0 · AWQ W4A164 contexts (0–56K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 56K
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
- AWQ 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. KV cache: fp8 # 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 Sohailhosseini/Fara1.5-27B-AWQ-W4A16 --max-model-len 65536 --gpu-memory-utilization 0.94 --max-num-seqs 51 --enable-prefix-caching --trust-remote-code --language-model-only --kv-cache-dtype fp8 --max-cudagraph-capture-size 16
Prefill
prompt tok/s by context length8K context1,8494,430 ms TTFT
16K context1,7909,064 ms TTFT
32K context1,64619,614 ms TTFT
56K context1,47438,260 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 | 46.970 ms TTFT | 87.1127 ms TTFT | 167.9209 ms TTFT | 326.6384 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 16K | Looped | 83.6546 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 32K | 44.2696 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
| 56K | Looped | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity | Over capacity |
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 | 1,849.0 | — | 4,430 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 1,790.0 | — | 9,064 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 1,646.0 | — | 19,614 ms | — | — | 1 | — | — |
| Prefill | 56K | 1 | Valid | 1,474.0 | — | 38,260 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 46.9 | 46.9 | 70 ms | 21.3 ms | 1.0 | 0 | 1,406 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 87.1 | 43.6 | 127 ms | 22.9 ms | 2.0 | 0 | 2,612 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 167.9 | 42.0 | 209 ms | 23.8 ms | 4.0 | 0 | 5,036 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 326.6 | 40.9 | 384 ms | 24.4 ms | 8.0 | 0 | 9,792 | 30.0 s |
| Sustained decode | 0 | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 16K | 2 | Valid | 83.6 | 41.6 | 546 ms | 24.0 ms | 2.0 | 0 | 2,508 | 30.0 s |
| Sustained decode | 16K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 1 | Valid | 44.2 | 44.2 | 696 ms | 22.6 ms | 1.0 | 0 | 1,324 | 30.0 s |
| Sustained decode | 32K | 2 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| 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 | 56K | 1 | Looped | — | — | — | — | — | 0 | 0 | — |
| Sustained decode | 56K | 2 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 56K | 128 | Over capacity | — | — | — | — | — | — | — | — |
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.