Speed test matrix

Qwen3.5-122B-A10B-GPTQ-Int4 on CMP 170HX 64GB ×2

vllm · 0.32.0rc1.dev249+g18305948a · GPTQ Int46 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 11:04 UTC

Hardware

Type: Discrete GPU
GPU: CMP 170HX
GPU count: 2
VRAM: 64 GB (2 × 32 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: GPTQ Int4
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: MTP, 2 tokens lookahead. Layout: PP2 (stock vLLM flag)
# 2x 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 Qwen/Qwen3.5-122B-A10B-GPTQ-Int4 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --pipeline-parallel-size 2 --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}

Prefill

prompt tok/s by context length

8K context5,4761,487 ms TTFT

16K context5,5492,904 ms TTFT

32K context5,1926,180 ms TTFT

64K context4,38714,594 ms TTFT

128K context3,26439,179 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	116.7137 ms TTFT	196.7184 ms TTFT	307.9249 ms TTFT	464.2362 ms TTFT	676.0526 ms TTFT	971.9773 ms TTFT	1,006.6Requests queued876 ms TTFT	Over capacity
16K	80.0964 ms TTFT	130.51,353 ms TTFT	197.02,319 ms TTFT	323.53,646 ms TTFT	416.16,162 ms TTFT	510.511,507 ms TTFT	Over capacity	Over capacity
32K	62.21,122 ms TTFT	102.71,569 ms TTFT	183.32,800 ms TTFT	236.53,885 ms TTFT	282.66,348 ms TTFT	Over capacity	Over capacity	Over capacity
64K	40.51,699 ms TTFT	72.02,285 ms TTFT	128.44,024 ms TTFT	128.85,766 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
128K	26.23,028 ms TTFT	47.03,978 ms TTFT	52.66,521 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	14.94,758 ms TTFT	26.76,508 ms TTFT	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	5,476.0	—	1,487 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,549.0	—	2,904 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,192.0	—	6,180 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,387.0	—	14,594 ms	—	—	1	—	—
Prefill	128K	1	Valid	3,264.0	—	39,179 ms	—	—	1	—	—
Sustained decode	0	1	Valid	116.7	117.7	137 ms	8.5 ms	1.0	0	3,502	30.0 s
Sustained decode	0	2	Valid	196.7	98.4	184 ms	10.2 ms	2.0	0	5,899	30.0 s
Sustained decode	0	4	Valid	307.9	77.1	249 ms	13.0 ms	4.0	0	9,235	30.0 s
Sustained decode	0	8	Valid	464.2	59.1	362 ms	16.9 ms	8.0	0	13,927	30.0 s
Sustained decode	0	16	Valid	676.0	43.7	526 ms	22.9 ms	16.0	0	20,281	30.0 s
Sustained decode	0	32	Valid	971.9	31.6	773 ms	31.6 ms	32.0	0	29,166	30.0 s
Sustained decode	0	64	Requests queued	1,006.6	23.0	876 ms	43.5 ms	52.5	0	30,193	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	80.0	81.0	964 ms	12.4 ms	1.0	0	2,399	30.0 s
Sustained decode	16K	2	Valid	130.5	65.5	1,353 ms	15.3 ms	2.0	0	3,914	30.0 s
Sustained decode	16K	4	Valid	197.0	48.8	2,319 ms	20.5 ms	4.0	0	5,909	30.0 s
Sustained decode	16K	8	Valid	323.5	38.9	3,646 ms	25.7 ms	8.0	0	9,702	30.0 s
Sustained decode	16K	16	Valid	416.1	23.9	6,162 ms	41.9 ms	16.0	0	12,478	30.0 s
Sustained decode	16K	32	Valid	510.5	13.4	11,507 ms	74.5 ms	32.0	0	15,281	29.9 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	62.2	63.2	1,122 ms	15.8 ms	1.0	0	1,866	30.0 s
Sustained decode	32K	2	Valid	102.7	51.5	1,569 ms	19.4 ms	2.0	0	3,079	30.0 s
Sustained decode	32K	4	Valid	183.3	45.7	2,800 ms	21.9 ms	4.0	0	5,500	30.0 s
Sustained decode	32K	8	Valid	236.5	29.0	3,885 ms	34.5 ms	8.0	0	7,087	30.0 s
Sustained decode	32K	16	Valid	282.6	16.2	6,348 ms	61.6 ms	16.0	0	8,474	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	40.5	41.4	1,699 ms	24.2 ms	1.0	0	1,215	30.0 s
Sustained decode	64K	2	Valid	72.0	35.8	2,285 ms	28.0 ms	2.0	0	2,159	30.0 s
Sustained decode	64K	4	Valid	128.4	32.0	4,024 ms	31.3 ms	4.0	0	3,849	30.0 s
Sustained decode	64K	8	Valid	128.8	15.5	5,766 ms	64.4 ms	8.0	0	3,861	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	128K	1	Valid	26.2	26.6	3,028 ms	37.6 ms	1.0	0	785	30.0 s
Sustained decode	128K	2	Valid	47.0	23.4	3,978 ms	42.8 ms	2.0	0	1,411	30.0 s
Sustained decode	128K	4	Valid	52.6	13.0	6,521 ms	77.0 ms	4.0	0	1,574	29.9 s
Sustained decode	128K	8	Over capacity	—	—	—	—	—	—	—	—
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	Valid	14.9	15.2	4,758 ms	65.7 ms	1.0	0	444	29.8 s
Sustained decode	248K	2	Valid	26.7	13.2	6,508 ms	75.7 ms	2.0	0	801	30.0 s
Sustained decode	248K	4	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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.