Speed test matrix

Muse-Glimmer-30B-FP8-block on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 block5 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 18:23 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
VRAM: 64 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: FP8 block
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
# 1x 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 RedHatAI/Muse-Glimmer-30B-FP8-block --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only

Prefill

prompt tok/s by context length

8K context2,2883,568 ms TTFT

16K context2,2137,281 ms TTFT

32K context2,12815,053 ms TTFT

64K context1,99132,091 ms TTFT

120K context1,79566,639 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	35.093 ms TTFT	69.0110 ms TTFT	135.0134 ms TTFT	253.3166 ms TTFT	462.2293 ms TTFT	640.7407 ms TTFT	955.2612 ms TTFT	1,279.71,053 ms TTFT
16K	33.1179 ms TTFT	65.0223 ms TTFT	122.7314 ms TTFT	213.7456 ms TTFT	348.8779 ms TTFT	431.41,362 ms TTFT	538.92,510 ms TTFT	Over capacity
32K	32.9268 ms TTFT	64.4383 ms TTFT	120.5584 ms TTFT	207.4887 ms TTFT	333.51,338 ms TTFT	408.72,451 ms TTFT	Over capacity	Over capacity
64K	32.4418 ms TTFT	63.2637 ms TTFT	116.6888 ms TTFT	195.71,450 ms TTFT	304.62,336 ms TTFT	Over capacity	Over capacity	Over capacity
120K	31.7875 ms TTFT	61.01,106 ms TTFT	109.71,498 ms TTFT	177.42,354 ms TTFT	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	2,288.0	—	3,568 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,213.0	—	7,281 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,128.0	—	15,053 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,991.0	—	32,091 ms	—	—	1	—	—
Prefill	120K	1	Valid	1,795.0	—	66,639 ms	—	—	1	—	—
Sustained decode	0	1	Valid	35.0	35.1	93 ms	28.5 ms	1.0	0	1,049	30.0 s
Sustained decode	0	2	Valid	69.0	34.6	110 ms	28.9 ms	2.0	0	2,068	30.0 s
Sustained decode	0	4	Valid	135.0	34.0	134 ms	29.4 ms	4.0	0	4,048	30.0 s
Sustained decode	0	8	Valid	253.3	32.0	166 ms	31.2 ms	8.0	0	7,592	30.0 s
Sustained decode	0	16	Valid	462.2	29.4	293 ms	34.0 ms	16.0	0	13,856	30.0 s
Sustained decode	0	32	Valid	640.7	20.4	407 ms	49.1 ms	32.0	0	19,200	30.0 s
Sustained decode	0	64	Valid	955.2	15.2	612 ms	65.6 ms	64.0	0	28,608	29.9 s
Sustained decode	0	128	Valid	1,279.7	10.2	1,053 ms	98.4 ms	128.0	0	38,400	30.0 s
Sustained decode	16K	1	Valid	33.1	37.5	179 ms	26.6 ms	1.0	0	993	30.0 s
Sustained decode	16K	2	Valid	65.0	35.1	223 ms	28.7 ms	2.0	0	1,948	30.0 s
Sustained decode	16K	4	Valid	122.7	30.7	314 ms	32.5 ms	4.0	0	3,680	30.0 s
Sustained decode	16K	8	Valid	213.7	26.8	456 ms	37.4 ms	8.0	0	6,408	30.0 s
Sustained decode	16K	16	Valid	348.8	21.9	779 ms	45.7 ms	16.0	0	10,464	30.0 s
Sustained decode	16K	32	Valid	431.4	13.5	1,362 ms	73.9 ms	32.0	0	12,928	30.0 s
Sustained decode	16K	64	Valid	538.9	8.6	2,510 ms	116.9 ms	64.0	0	16,128	29.9 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	32.9	42.1	268 ms	23.8 ms	1.0	0	985	30.0 s
Sustained decode	32K	2	Valid	64.4	32.3	383 ms	31.0 ms	2.0	0	1,932	30.0 s
Sustained decode	32K	4	Valid	120.5	30.2	584 ms	33.1 ms	4.0	0	3,616	30.0 s
Sustained decode	32K	8	Valid	207.4	25.9	887 ms	38.6 ms	8.0	0	6,216	30.0 s
Sustained decode	32K	16	Valid	333.5	20.7	1,338 ms	48.4 ms	16.0	0	10,000	30.0 s
Sustained decode	32K	32	Valid	408.7	12.6	2,451 ms	79.5 ms	32.0	0	12,256	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	32.4	44.9	418 ms	22.3 ms	1.0	0	973	30.0 s
Sustained decode	64K	2	Valid	63.2	37.0	637 ms	27.6 ms	2.0	0	1,894	30.0 s
Sustained decode	64K	4	Valid	116.6	29.2	888 ms	34.2 ms	4.0	0	3,496	30.0 s
Sustained decode	64K	8	Valid	195.7	24.4	1,450 ms	40.9 ms	8.0	0	5,864	30.0 s
Sustained decode	64K	16	Valid	304.6	18.8	2,336 ms	53.1 ms	16.0	0	9,136	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	1	Valid	31.7	31.8	875 ms	31.4 ms	1.0	0	950	30.0 s
Sustained decode	120K	2	Valid	61.0	43.7	1,106 ms	23.0 ms	2.0	0	1,830	30.0 s
Sustained decode	120K	4	Valid	109.7	30.6	1,498 ms	33.0 ms	4.0	0	3,288	30.0 s
Sustained decode	120K	8	Valid	177.4	22.1	2,354 ms	45.3 ms	8.0	0	5,320	30.0 s
Sustained decode	120K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	128	Over capacity	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.