Speed test matrix

gemma-3-12b-it-FP8-dynamic on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP8 dynamic5 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, 08:11 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-3-12b-it-FP8-dynamic --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 context4,8821,632 ms TTFT

16K context4,7103,348 ms TTFT

32K context4,5666,875 ms TTFT

64K context4,15315,085 ms TTFT

120K context3,54833,070 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	70.969 ms TTFT	137.297 ms TTFT	263.2105 ms TTFT	441.9145 ms TTFT	686.9193 ms TTFT	936.8335 ms TTFT	1,335.5528 ms TTFT	Over capacity
16K	65.7221 ms TTFT	120.5295 ms TTFT	213.1384 ms TTFT	322.1717 ms TTFT	439.81,261 ms TTFT	Over capacity	Over capacity	Over capacity
32K	61.8324 ms TTFT	108.3431 ms TTFT	180.6663 ms TTFT	255.91,020 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	55.4631 ms TTFT	90.1879 ms TTFT	138.01,347 ms TTFT	181.32,085 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
120K	46.71,212 ms TTFT	69.61,777 ms TTFT	97.82,631 ms TTFT	Over capacity	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	4,882.0	—	1,632 ms	—	—	1	—	—
Prefill	16K	1	Valid	4,710.0	—	3,348 ms	—	—	1	—	—
Prefill	32K	1	Valid	4,566.0	—	6,875 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,153.0	—	15,085 ms	—	—	1	—	—
Prefill	120K	1	Valid	3,548.0	—	33,070 ms	—	—	1	—	—
Sustained decode	0	1	Valid	70.9	81.0	69 ms	12.4 ms	1.0	0	2,126	30.0 s
Sustained decode	0	2	Valid	137.2	72.4	97 ms	13.8 ms	2.0	0	4,116	30.0 s
Sustained decode	0	4	Valid	263.2	67.3	105 ms	14.8 ms	4.0	0	7,896	30.0 s
Sustained decode	0	8	Valid	441.9	57.5	145 ms	17.4 ms	8.0	0	13,256	30.0 s
Sustained decode	0	16	Valid	686.9	45.9	193 ms	21.8 ms	16.0	0	20,624	30.0 s
Sustained decode	0	32	Valid	936.8	31.5	335 ms	31.7 ms	32.0	0	28,096	30.0 s
Sustained decode	0	64	Valid	1,335.5	22.7	528 ms	44.1 ms	64.0	0	40,064	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	65.7	65.9	221 ms	15.2 ms	1.0	0	1,971	30.0 s
Sustained decode	16K	2	Valid	120.5	63.3	295 ms	15.8 ms	2.0	0	3,616	30.0 s
Sustained decode	16K	4	Valid	213.1	53.5	384 ms	18.7 ms	4.0	0	6,392	30.0 s
Sustained decode	16K	8	Valid	322.1	40.5	717 ms	24.7 ms	8.0	0	9,656	30.0 s
Sustained decode	16K	16	Valid	439.8	27.6	1,261 ms	36.2 ms	16.0	0	13,184	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	61.8	62.0	324 ms	16.1 ms	1.0	0	1,853	30.0 s
Sustained decode	32K	2	Valid	108.3	54.3	431 ms	18.4 ms	2.0	0	3,246	30.0 s
Sustained decode	32K	4	Valid	180.6	45.3	663 ms	22.1 ms	4.0	0	5,416	30.0 s
Sustained decode	32K	8	Valid	255.9	32.1	1,020 ms	31.1 ms	8.0	0	7,672	30.0 s
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	Valid	55.4	55.6	631 ms	18.0 ms	1.0	0	1,661	30.0 s
Sustained decode	64K	2	Valid	90.1	45.3	879 ms	22.1 ms	2.0	0	2,702	30.0 s
Sustained decode	64K	4	Valid	138.0	34.7	1,347 ms	28.8 ms	4.0	0	4,141	30.0 s
Sustained decode	64K	8	Valid	181.3	22.8	2,085 ms	43.8 ms	8.0	0	5,440	30.0 s
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	120K	1	Valid	46.7	47.0	1,212 ms	21.3 ms	1.0	0	1,400	30.0 s
Sustained decode	120K	2	Valid	69.6	35.1	1,777 ms	28.5 ms	2.0	0	2,086	30.0 s
Sustained decode	120K	4	Valid	97.8	24.7	2,631 ms	40.4 ms	4.0	0	2,932	30.0 s
Sustained decode	120K	8	Over capacity	—	—	—	—	—	—	—	—
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.