Speed test matrix
internlm_internlm3-8b-instruct-autoround-int8-gs128-sym on RTX 3090 24GB
vllm · 0.32.0 · AutoRound INT8 W8A16 gs1283 contexts (0–24K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 24K
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
- AutoRound INT8 W8A16 gs128
- 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 fbaldassarri/internlm_internlm3-8b-instruct-autoround-int8-gs128-sym --max-model-len 32768 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --hf-overrides {"quantization_config":{"quant_method":"gptq","bits":8,"group_size":128,"sym":true,"desc_act":false,"lm_head":false}}Prefill
prompt tok/s by context length8K context4,9612,021 ms TTFT
16K context4,1584,780 ms TTFT
24K context3,5608,346 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 | 79.833 ms TTFT | 159.029 ms TTFT | 313.136 ms TTFT | 604.043 ms TTFT | 1,127.162 ms TTFT | Over capacity | Over capacity | Over capacity |
| 16K | 71.971 ms TTFT | 135.295 ms TTFT | 254.1166 ms TTFT | 431.9259 ms TTFT | Over capacity | Over capacity | Over capacity | Over capacity |
| 24K | 67.9129 ms TTFT | 127.3166 ms TTFT | 232.5254 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 | 4,961.0 | — | 2,021 ms | — | — | 1 | — | — |
| Prefill | 16K | 1 | Valid | 4,158.0 | — | 4,780 ms | — | — | 1 | — | — |
| Prefill | 24K | 1 | Valid | 3,560.0 | — | 8,346 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 79.8 | 79.9 | 33 ms | 12.5 ms | 1.0 | 0 | 2,395 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 159.0 | 79.7 | 29 ms | 12.5 ms | 2.0 | 0 | 4,770 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 313.1 | 78.6 | 36 ms | 12.7 ms | 4.0 | 0 | 9,392 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 604.0 | 76.2 | 43 ms | 13.1 ms | 8.0 | 0 | 18,120 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 1,127.1 | 71.5 | 62 ms | 14.0 ms | 16.0 | 0 | 33,808 | 30.0 s |
| Sustained decode | 0 | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 0 | 128 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 16K | 1 | Valid | 71.9 | 72.0 | 71 ms | 13.9 ms | 1.0 | 0 | 2,157 | 30.0 s |
| Sustained decode | 16K | 2 | Valid | 135.2 | 67.7 | 95 ms | 14.8 ms | 2.0 | 0 | 4,056 | 30.0 s |
| Sustained decode | 16K | 4 | Valid | 254.1 | 63.7 | 166 ms | 15.7 ms | 4.0 | 0 | 7,620 | 30.0 s |
| Sustained decode | 16K | 8 | Valid | 431.9 | 54.1 | 259 ms | 18.5 ms | 8.0 | 0 | 12,952 | 30.0 s |
| 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 | 24K | 1 | Valid | 67.9 | 68.0 | 129 ms | 14.7 ms | 1.0 | 0 | 2,037 | 30.0 s |
| Sustained decode | 24K | 2 | Valid | 127.3 | 63.6 | 166 ms | 15.7 ms | 2.0 | 0 | 3,818 | 30.0 s |
| Sustained decode | 24K | 4 | Valid | 232.5 | 58.1 | 254 ms | 17.2 ms | 4.0 | 0 | 6,972 | 30.0 s |
| Sustained decode | 24K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 32 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 64 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 24K | 128 | Over capacity | — | — | — | — | — | — | — | — |
Over capacity: not run because this many requests at this context would exceed the server's KV cache.