Speed test matrix
GLM-4.7-Flash on Intel Arc Pro B65 32GB ×4
vllm · 0.32.0 · BF165 contexts (0–128K) × 5 concurrency levels (1–16) · prefill at 8K, 32K, 64K, 128K
Hardware
- Type
- Discrete GPU
- GPU
- Intel Arc Pro B65
- GPU count
- 4
- VRAM
- 128 GB (4 × 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
- BF16
- Build
- vllm/vllm-openai-xpu:v0.32.0@sha256:20509fabe50dda62c09c0786d1a6effa299d3dbdddcc49331b93e65144652a8c
Run
- Served model
- glm47
- Revision
- 7dd20894a642a0aa287e9827cb1a1f7f91386b67
- 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-glm47-u-tp4-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-a/glm47-u-tp4-ctx/session:/bench -e ZE_AFFINITY_MASK=0\,1\,2\,3 -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/GLM-4.7-Flash --host 0.0.0.0 --served-model-name glm47 --tensor-parallel-size 4 --pipeline-parallel-size 1 --data-parallel-size 1 --gpu-memory-utilization 0.92 --max-model-len 202752 --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 context9,041907 ms TTFT
32K context6,1735,239 ms TTFT
64K context3,98116,215 ms TTFT
128K context2,34954,899 ms TTFT
Sustained decode
total tok/s · rows are context length, columns are simultaneous requests| Context | 1 | 2 | 4 | 8 | 16 |
|---|---|---|---|---|---|
| 0 | 81.039 ms TTFT | 142.271 ms TTFT | 241.369 ms TTFT | 377.8147 ms TTFT | 595.7208 ms TTFT |
| 8K | 64.3125 ms TTFT | 104.7183 ms TTFT | 166.0272 ms TTFT | 238.4420 ms TTFT | 332.4625 ms TTFT |
| 32K | 43.8420 ms TTFT | 66.0583 ms TTFT | 92.11,153 ms TTFT | Over capacity | Over capacity |
| 64K | 31.6789 ms TTFT | 46.8Requests queued1,110 ms TTFT | Over capacity | Over capacity | Over capacity |
| 128K | 21.01,603 ms TTFT | 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. 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 | 9,041.0 | — | 907 ms | — | — | 1 | — | — |
| Prefill | 32K | 1 | Valid | 6,173.0 | — | 5,239 ms | — | — | 1 | — | — |
| Prefill | 64K | 1 | Valid | 3,981.0 | — | 16,215 ms | — | — | 1 | — | — |
| Prefill | 128K | 1 | Valid | 2,349.0 | — | 54,899 ms | — | — | 1 | — | — |
| Sustained decode | 0 | 1 | Valid | 81.0 | 81.8 | 39 ms | 12.2 ms | 0.9 | 11 | 2,430 | 30.0 s |
| Sustained decode | 0 | 2 | Valid | 142.2 | 72.0 | 71 ms | 13.9 ms | 2.0 | 18 | 4,265 | 30.0 s |
| Sustained decode | 0 | 4 | Valid | 241.3 | 61.5 | 69 ms | 16.3 ms | 4.0 | 32 | 7,237 | 30.0 s |
| Sustained decode | 0 | 8 | Valid | 377.8 | 48.2 | 147 ms | 20.7 ms | 7.9 | 48 | 11,327 | 30.0 s |
| Sustained decode | 0 | 16 | Valid | 595.7 | 38.3 | 208 ms | 26.1 ms | 15.8 | 80 | 17,869 | 30.0 s |
| Sustained decode | 8K | 1 | Valid | 64.3 | 66.3 | 125 ms | 15.1 ms | 1.0 | 8 | 1,930 | 30.0 s |
| Sustained decode | 8K | 2 | Valid | 104.7 | 54.6 | 183 ms | 18.3 ms | 2.0 | 14 | 3,140 | 30.0 s |
| Sustained decode | 8K | 4 | Valid | 166.0 | 43.2 | 272 ms | 23.2 ms | 3.9 | 20 | 4,980 | 30.0 s |
| Sustained decode | 8K | 8 | Valid | 238.4 | 31.6 | 420 ms | 31.7 ms | 7.7 | 32 | 7,126 | 29.9 s |
| Sustained decode | 8K | 16 | Valid | 332.4 | 21.5 | 625 ms | 46.4 ms | 15.8 | 32 | 9,963 | 30.0 s |
| Sustained decode | 32K | 1 | Valid | 43.8 | 46.9 | 420 ms | 21.3 ms | 0.9 | 5 | 1,314 | 30.0 s |
| Sustained decode | 32K | 2 | Valid | 66.0 | 35.7 | 583 ms | 28.0 ms | 1.9 | 8 | 1,980 | 30.0 s |
| Sustained decode | 32K | 4 | Valid | 92.1 | 24.9 | 1,153 ms | 40.1 ms | 3.9 | 12 | 2,734 | 29.7 s |
| Sustained decode | 32K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 32K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 1 | Valid | 31.6 | 35.1 | 789 ms | 28.5 ms | 0.9 | 4 | 947 | 30.0 s |
| Sustained decode | 64K | 2 | Requests queued | 46.8 | 25.1 | 1,110 ms | 39.9 ms | 1.9 | 6 | 1,374 | 29.4 s |
| Sustained decode | 64K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 64K | 16 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 1 | Valid | 21.0 | 23.4 | 1,603 ms | 42.6 ms | 0.9 | 2 | 629 | 30.0 s |
| Sustained decode | 128K | 2 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 4 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 8 | Over capacity | — | — | — | — | — | — | — | — |
| Sustained decode | 128K | 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.