Speed test matrix

GLM-4.7-185B-W4A16 on CMP 170HX 64GB ×2

vllm · 0.32.0rc1.dev249+g18305948a · GPTQ W4A166 contexts (0–184K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 10:06 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
GPU count: 2
VRAM: 64 GB (2 × 32 GB)
CPU: 2x Intel Xeon E5-2630 v3
RAM: 64 GB
OS: Ubuntu 26.04.1

Engine

Engine: vllm
Version: 0.32.0rc1.dev249+g18305948a
Quantization: GPTQ W4A16
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: ghcr.io/morrowmake/vllm-cmp170hx@sha256:5b45c4d6feafefe2eba962375b0a9a0b64fd22b9e05abb4d1966f14a32a2db79

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: int4_per_token_head. Layout: PP2 (stock vLLM flag)
# 2x NVIDIA CMP 170HX 64GB (cmpunlocker), 180 W limit, PCIe Gen2 x4 | 2x Intel Xeon E5-2630 v3 | 64 GB DDR4 | Ubuntu 26.04.1, NVIDIA driver 610.57.04 | vLLM: Morrowmake vllm-cmp170hx image
vllm serve 0xSero/GLM-4.7-185B-W4A16 --max-model-len 202752 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --pipeline-parallel-size 2 --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context596.013,721 ms TTFT

16K context407.039,809 ms TTFT

32K context245.0131,487 ms TTFT

64K context136.0473,216 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	15.4Requests queued76,465 ms TTFT	45.0Requests queued63,655 ms TTFT	88.5358 ms TTFT	120.1397 ms TTFT	183.9475 ms TTFT	Over capacity	Over capacity	Over capacity
16K	20.5453 ms TTFT	37.8562 ms TTFT	44.0899 ms TTFT	60.61,024 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	14.7Requests queued697 ms TTFT	27.2Requests queued891 ms TTFT	31.71,539 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	9.3Requests queued1,094 ms TTFT	17.3Requests queued1,496 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
184K	Over capacity	Over capacity	Over capacity	Over capacity	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	596.0	—	13,721 ms	—	—	1	—	—
Prefill	16K	1	Valid	407.0	—	39,809 ms	—	—	1	—	—
Prefill	32K	1	Valid	245.0	—	131,487 ms	—	—	1	—	—
Prefill	64K	1	Valid	136.0	—	473,216 ms	—	—	1	—	—
Sustained decode	0	1	Requests queued	15.4	33.9	76,465 ms	29.5 ms	1.0	0	461	30.0 s
Sustained decode	0	2	Requests queued	45.0	25.6	63,655 ms	39.1 ms	1.7	0	1,350	30.0 s
Sustained decode	0	4	Valid	88.5	22.2	358 ms	45.0 ms	4.0	0	2,656	30.0 s
Sustained decode	0	8	Valid	120.1	15.2	397 ms	65.7 ms	8.0	0	3,600	30.0 s
Sustained decode	0	16	Valid	183.9	11.8	475 ms	84.9 ms	16.0	0	5,505	29.9 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	20.5	20.5	453 ms	48.8 ms	1.0	0	614	30.0 s
Sustained decode	16K	2	Valid	37.8	18.9	562 ms	52.9 ms	2.0	0	1,135	30.0 s
Sustained decode	16K	4	Valid	44.0	11.1	899 ms	90.5 ms	4.0	0	1,320	30.0 s
Sustained decode	16K	8	Valid	60.6	7.6	1,024 ms	132.4 ms	8.0	0	1,816	29.9 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	32K	1	Requests queued	14.7	14.7	697 ms	67.9 ms	1.0	0	440	30.0 s
Sustained decode	32K	2	Requests queued	27.2	13.8	891 ms	72.6 ms	1.9	0	816	30.0 s
Sustained decode	32K	4	Valid	31.7	8.0	1,539 ms	125.1 ms	4.0	0	949	30.0 s
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	64K	1	Requests queued	9.3	9.5	1,094 ms	105.7 ms	1.0	0	279	30.0 s
Sustained decode	64K	2	Requests queued	17.3	8.9	1,496 ms	112.1 ms	1.9	0	519	30.0 s
Sustained decode	64K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Over capacity	—	—	—	—	0.4	0	0	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	—	—	—	—	—	—	—	—
Sustained decode	128K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	1	Over capacity	—	—	—	—	0.4	0	0	30.0 s
Sustained decode	184K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	184K	128	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.