Speed test matrix

granite-4.2-3b-fp8 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · FP85 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, 09:39 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
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 ibm-granite/granite-4.2-3b-fp8 --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context11.4K718 ms TTFT

16K context9,7701,659 ms TTFT

32K context7,4264,345 ms TTFT

64K context4,99512,888 ms TTFT

120K context3,15338,230 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	177.564 ms TTFT	330.358 ms TTFT	611.960 ms TTFT	1,078.698 ms TTFT	1,792.5141 ms TTFT	2,585.5224 ms TTFT	3,789.9307 ms TTFT	Over capacity
16K	146.9172 ms TTFT	251.5212 ms TTFT	399.1287 ms TTFT	571.6462 ms TTFT	709.9653 ms TTFT	Over capacity	Over capacity	Over capacity
32K	125.1237 ms TTFT	198.5370 ms TTFT	286.7517 ms TTFT	367.7722 ms TTFT	421.31,407 ms TTFT	Over capacity	Over capacity	Over capacity
64K	96.0416 ms TTFT	137.4626 ms TTFT	177.9891 ms TTFT	209.21,425 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
120K	68.9841 ms TTFT	86.51,113 ms TTFT	105.41,506 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	11.4K	—	718 ms	—	—	1	—	—
Prefill	16K	1	Valid	9,770.0	—	1,659 ms	—	—	1	—	—
Prefill	32K	1	Valid	7,426.0	—	4,345 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,995.0	—	12,888 ms	—	—	1	—	—
Prefill	120K	1	Valid	3,153.0	—	38,230 ms	—	—	1	—	—
Sustained decode	0	1	Valid	177.5	179.2	64 ms	5.6 ms	1.0	0	5,325	30.0 s
Sustained decode	0	2	Valid	330.3	167.6	58 ms	6.0 ms	2.0	0	9,910	30.0 s
Sustained decode	0	4	Valid	611.9	156.9	60 ms	6.4 ms	4.0	0	18,373	30.0 s
Sustained decode	0	8	Valid	1,078.6	140.4	98 ms	7.1 ms	8.0	0	32,352	30.0 s
Sustained decode	0	16	Valid	1,792.5	120.2	141 ms	8.3 ms	16.0	0	53,767	30.0 s
Sustained decode	0	32	Valid	2,585.5	87.5	224 ms	11.4 ms	32.0	0	77,600	30.0 s
Sustained decode	0	64	Valid	3,789.9	64.7	307 ms	15.5 ms	64.0	0	113,728	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	146.9	147.5	172 ms	6.8 ms	1.0	0	4,406	30.0 s
Sustained decode	16K	2	Valid	251.5	126.5	212 ms	7.9 ms	2.0	0	7,546	30.0 s
Sustained decode	16K	4	Valid	399.1	100.5	287 ms	10.0 ms	4.0	0	11,972	30.0 s
Sustained decode	16K	8	Valid	571.6	71.8	462 ms	13.9 ms	8.0	0	17,144	30.0 s
Sustained decode	16K	16	Valid	709.9	44.3	653 ms	22.6 ms	16.0	0	21,296	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	125.1	125.8	237 ms	8.0 ms	1.0	0	3,755	30.0 s
Sustained decode	32K	2	Valid	198.5	99.8	370 ms	10.0 ms	2.0	0	5,954	30.0 s
Sustained decode	32K	4	Valid	286.7	72.0	517 ms	13.9 ms	4.0	0	8,600	30.0 s
Sustained decode	32K	8	Valid	367.7	45.9	722 ms	21.8 ms	8.0	0	11,032	30.0 s
Sustained decode	32K	16	Valid	421.3	26.1	1,407 ms	38.3 ms	16.0	0	12,628	30.0 s
Sustained decode	32K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	96.0	96.4	416 ms	10.4 ms	1.0	0	2,879	30.0 s
Sustained decode	64K	2	Valid	137.4	69.0	626 ms	14.5 ms	2.0	0	4,120	30.0 s
Sustained decode	64K	4	Valid	177.9	44.6	891 ms	22.4 ms	4.0	0	5,336	30.0 s
Sustained decode	64K	8	Valid	209.2	26.1	1,425 ms	38.3 ms	8.0	0	6,272	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	68.9	69.2	841 ms	14.5 ms	1.0	0	2,066	30.0 s
Sustained decode	120K	2	Valid	86.5	43.5	1,113 ms	23.0 ms	2.0	0	2,594	30.0 s
Sustained decode	120K	4	Valid	105.4	26.4	1,506 ms	37.8 ms	4.0	0	3,160	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.