Speed test matrix

gemma-4-26B-A4B-it-FP8-dynamic on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 dynamic6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 06:04 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 dynamic
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/gemma-4-26B-A4B-it-FP8-dynamic --max-model-len 262144 --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 context7,2651,097 ms TTFT

16K context5,9372,657 ms TTFT

32K context4,5626,882 ms TTFT

64K context3,04920,548 ms TTFT

128K context1,83368,274 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	128.572 ms TTFT	220.392 ms TTFT	Looped	Looped	Looped	1,317.2279 ms TTFT	1,926.3563 ms TTFT	2,919.3839 ms TTFT
16K	102.0236 ms TTFT	178.9305 ms TTFT	Looped	Looped	Looped	Looped	Over capacity	Over capacity
32K	91.8335 ms TTFT	Looped	Looped	Looped	Looped	Looped	Over capacity	Over capacity
64K	80.9628 ms TTFT	Looped	Looped	Looped	Looped	Over capacity	Over capacity	Over capacity
128K	Looped	Looped	Looped	Looped	Over capacity	Over capacity	Over capacity	Over capacity
248K	Looped	Looped	Looped	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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	7,265.0	—	1,097 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,937.0	—	2,657 ms	—	—	1	—	—
Prefill	32K	1	Valid	4,562.0	—	6,882 ms	—	—	1	—	—
Prefill	64K	1	Valid	3,049.0	—	20,548 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,833.0	—	68,274 ms	—	—	1	—	—
Sustained decode	0	1	Valid	128.5	129.7	72 ms	7.7 ms	1.0	0	3,856	30.0 s
Sustained decode	0	2	Valid	220.3	111.3	92 ms	9.0 ms	2.0	0	6,610	30.0 s
Sustained decode	0	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	0	32	Valid	1,317.2	43.1	279 ms	23.2 ms	32.0	0	39,520	30.0 s
Sustained decode	0	64	Valid	1,926.3	31.9	563 ms	31.4 ms	64.0	0	57,792	30.0 s
Sustained decode	0	128	Valid	2,919.3	24.2	839 ms	41.3 ms	128.0	0	87,680	30.0 s
Sustained decode	16K	1	Valid	102.0	102.6	236 ms	9.7 ms	1.0	0	3,060	30.0 s
Sustained decode	16K	2	Valid	178.9	90.1	305 ms	11.1 ms	2.0	0	5,366	30.0 s
Sustained decode	16K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	32	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	91.8	92.1	335 ms	10.9 ms	1.0	0	2,755	30.0 s
Sustained decode	32K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	32	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	80.9	3467.8	628 ms	0.3 ms	1.0	0	2,427	30.0 s
Sustained decode	64K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Looped	—	—	—	—	—	0	0	—
Sustained decode	128K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	128K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	128K	8	Looped	—	—	—	—	—	0	0	—
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	248K	1	Looped	—	—	—	—	—	0	0	—
Sustained decode	248K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	248K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	248K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	248K	128	Over capacity	—	—	—	—	—	—	—	—

Looped: the model kept repeating the same text, so the benchmark discarded this measurement. Over capacity: not run because this many requests at this context would exceed the server's KV cache.