Speed test matrix

Llama-3.3-70B-Instruct-NVFP4 on CMP 170HX 64GB

vllm · 0.32.0rc1.dev249+g18305948a · NVFP45 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 17:16 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: NVFP4
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
# 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 nvidia/Llama-3.3-70B-Instruct-NVFP4 --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code  --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context477.017,191 ms TTFT

16K context348.046,667 ms TTFT

32K context233.0138,418 ms TTFT

64K context140.0459,218 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	20.29,187 ms TTFT	39.96,342 ms TTFT	77.9395 ms TTFT	148.8414 ms TTFT	245.8434 ms TTFT	Over capacity	Over capacity	Over capacity
16K	15.7314 ms TTFT	29.0467 ms TTFT	46.3637 ms TTFT	66.2934 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	12.8Requests queued511 ms TTFT	22.7Requests queued745 ms TTFT	32.71,009 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	9.3Requests queued532 ms TTFT	15.7Requests queued766 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	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	477.0	—	17,191 ms	—	—	1	—	—
Prefill	16K	1	Valid	348.0	—	46,667 ms	—	—	1	—	—
Prefill	32K	1	Valid	233.0	—	138,418 ms	—	—	1	—	—
Prefill	64K	1	Valid	140.0	—	459,218 ms	—	—	1	—	—
Sustained decode	0	1	Valid	20.2	20.2	9,187 ms	49.5 ms	1.0	0	605	30.0 s
Sustained decode	0	2	Valid	39.9	20.0	6,342 ms	50.1 ms	2.0	0	1,196	30.0 s
Sustained decode	0	4	Valid	77.9	19.5	395 ms	51.2 ms	4.0	0	2,336	30.0 s
Sustained decode	0	8	Valid	148.8	18.7	414 ms	53.4 ms	8.0	0	4,464	30.0 s
Sustained decode	0	16	Valid	245.8	15.5	434 ms	64.6 ms	16.0	0	7,376	30.0 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	15.7	15.7	314 ms	63.9 ms	1.0	0	469	29.9 s
Sustained decode	16K	2	Valid	29.0	14.5	467 ms	69.0 ms	2.0	0	868	30.0 s
Sustained decode	16K	4	Valid	46.3	11.6	637 ms	86.1 ms	4.0	0	1,388	29.9 s
Sustained decode	16K	8	Valid	66.2	8.3	934 ms	120.7 ms	8.0	0	1,984	30.0 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	12.8	12.8	511 ms	78.4 ms	1.0	0	382	30.0 s
Sustained decode	32K	2	Requests queued	22.7	11.4	745 ms	88.1 ms	2.0	0	680	30.0 s
Sustained decode	32K	4	Valid	32.7	8.2	1,009 ms	122.0 ms	4.0	0	980	29.9 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	9.3	532 ms	107.7 ms	1.0	0	278	30.0 s
Sustained decode	64K	2	Requests queued	15.7	7.9	766 ms	126.5 ms	2.0	0	471	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	120K	1	Over capacity	—	—	—	—	0.3	0	0	30.0 s
Sustained decode	120K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	120K	4	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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.