Speed test matrix
gpt-oss-20b on Intel Arc Pro B65 32GB
vllm · 0.32.0 · MXFP44 contexts (0–64K) × 5 concurrency levels (1–16) · prefill at 8K, 32K, 64K
Hardware
- Type
- Discrete GPU
- GPU
- Intel Arc Pro B65
- VRAM
- 32 GB
- CPU
- AMD EPYC 7343 16-Core Processor
- RAM
- 499 GB
- OS
- Ubuntu 26.04.1 LTS
Engine
- Engine
- vllm
- Version
- 0.32.0
- Quantization
- MXFP4
- Build
- vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c
Run
- Served model
- gpt-oss
- Revision
- 6cee5e81ee83917806bbde320786a8fb61efebee
- 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
- 256 tokens
Launch command
docker run --rm -d --name lsv-oss20-u-tp1-ctx --net=host --shm-size=32g --device /dev/dri:/dev/dri -v /dev/dri:/dev/dri -v /models/weights:/llm/models:ro -v /models/bench/lmx-ctx-20261019/session-smoke/oss20-u-tp1-ctx/session:/bench -e ZE_AFFINITY_MASK=0 -e VLLM_XPU_ENABLE_XPU_GRAPH=1 -e DISABLE_ESIMD_GDN_OUTPROJ=0 -e DISABLE_ESIMD_INT4=0 -e DISABLE_ESIMD_INT4_GEMM=0 -e VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=1 -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 -e VLLM_WORKER_MULTIPROC_METHOD=spawn -e CCL_ZE_IPC_EXCHANGE=sockets --entrypoint vllm vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c serve --model /llm/models/gpt-oss-20b --host 0.0.0.0 --served-model-name gpt-oss --tensor-parallel-size 1 --pipeline-parallel-size 1 --data-parallel-size 1 --gpu-memory-utilization 0.92 --max-model-len 131072 --max-num-seqs 16 --max-num-batched-tokens 8192 --block-size 64 --kv-cache-dtype auto --mamba-ssm-cache-dtype auto --dtype bfloat16 --enable-prefix-caching --trust-remote-code --cpu-offload-gb 0 --port 8000
Prefill
prompt tok/s by context length8K context6,3261,294 ms TTFT
32K context4,9586,480 ms TTFT
64K context3,79816,865 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 |
|---|---|---|---|---|---|
| 0 | 97.357 ms TTFT | 171.666 ms TTFT | 279.290 ms TTFT | 472.9119 ms TTFT | 760.9171 ms TTFT |
| 8K | 91.8101 ms TTFT | 162.7137 ms TTFT | 260.4178 ms TTFT | 432.1Requests queued252 ms TTFT | 654.7Requests queued421 ms TTFT |
| 32K | 82.9121 ms TTFT | 148.6122 ms TTFT | 233.6162 ms TTFT | 371.5206 ms TTFT | Over capacity |
| 64K | 72.1215 ms TTFT | 129.3276 ms TTFT | 202.7395 ms TTFT | 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 | 6,326.0 | — | 1,294 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 4,958.0 | — | 6,480 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 3,798.0 | — | 16,865 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 97.3 | 99.1 | 57 ms | 10.1 ms | 1.0 | 13 | 2,918 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 171.6 | 87.8 | 66 ms | 11.4 ms | 2.0 | 22 | 5,146 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 279.2 | 71.9 | 90 ms | 13.9 ms | 3.9 | 36 | 8,375 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 472.9 | 61.1 | 119 ms | 16.4 ms | 8.0 | 64 | 14,179 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 760.9 | 49.2 | 171 ms | 20.3 ms | 15.6 | 96 | 22,815 | 30.0 s |
| Sustained decode | 8K | 1 | Valid | 91.8 | 94.8 | 101 ms | 10.6 ms | 0.9 | 12 | 2,751 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 162.7 | 84.8 | 137 ms | 11.8 ms | 2.0 | 20 | 4,881 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 260.4 | 69.0 | 178 ms | 14.5 ms | 4.0 | 32 | 7,813 | 30.0 s |
| Sustained decode | 8K | 8 | Requests queued | 432.1 | 56.2 | 252 ms | 17.8 ms | 7.5 | 56 | 12,962 | 30.0 s |
| Sustained decode | 8K | 16 | Requests queued | 654.7 | 44.2 | 421 ms | 22.6 ms | 15.1 | 80 | 19,640 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 82.9 | 87.2 | 121 ms | 11.5 ms | 1.0 | 10 | 2,486 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 148.6 | 76.8 | 122 ms | 13.0 ms | 2.0 | 19 | 4,457 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 233.6 | 62.3 | 162 ms | 16.1 ms | 4.0 | 29 | 7,008 | 30.0 s |
| Sustained decode | 32K | 8 | Valid | 371.5 | 48.7 | 206 ms | 20.5 ms | 7.8 | 48 | 11,141 | 30.0 s |
| Sustained decode | 32K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 72.1 | 77.3 | 215 ms | 12.9 ms | 0.9 | 9 | 2,164 | 30.0 s |
| Sustained decode | 64K | 2 | Valid | 129.3 | 69.8 | 276 ms | 14.3 ms | 2.0 | 16 | 3,878 | 30.0 s |
| Sustained decode | 64K | 4 | Valid | 202.7 | 55.0 | 395 ms | 18.2 ms | 4.0 | 24 | 6,079 | 30.0 s |
| Sustained decode | 64K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 16 | 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.