Speed test matrix

Qwen3.5-122B-A10B-AWQ-4bit on CMP 170HX 64GB ×2

vllm · 0.32.0rc1.dev249+g18305948a · AWQ 4-bit6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Nov 1, 2026, 19:00 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: AWQ 4-bit
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. 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 cyankiwi/Qwen3.5-122B-A10B-AWQ-4bit --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 --hf-overrides {"text_config":{"tie_word_embeddings":false}}

Prefill

prompt tok/s by context length

8K context5,0961,598 ms TTFT

16K context5,6592,848 ms TTFT

32K context5,3915,951 ms TTFT

64K context4,64113,796 ms TTFT

128K context3,52336,301 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	66.7126 ms TTFT	121.3162 ms TTFT	207.8226 ms TTFT	319.8324 ms TTFT	495.2504 ms TTFT	748.5738 ms TTFT	1,110.91,117 ms TTFT	1,613.31,515 ms TTFT
16K	65.0619 ms TTFT	111.4854 ms TTFT	198.51,269 ms TTFT	332.72,079 ms TTFT	499.83,679 ms TTFT	696.27,069 ms TTFT	948.614,094 ms TTFT	Over capacity
32K	63.3510 ms TTFT	108.6684 ms TTFT	192.5944 ms TTFT	315.51,518 ms TTFT	465.32,643 ms TTFT	632.35,039 ms TTFT	Over capacity	Over capacity
64K	60.91,026 ms TTFT	104.51,321 ms TTFT	181.31,872 ms TTFT	288.42,953 ms TTFT	408.05,187 ms TTFT	Over capacity	Over capacity	Over capacity
128K	56.9932 ms TTFT	97.81,014 ms TTFT	161.51,550 ms TTFT	244.52,394 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
248K	50.62,056 ms TTFT	86.82,538 ms TTFT	108.53,665 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	5,096.0	—	1,598 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,659.0	—	2,848 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,391.0	—	5,951 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,641.0	—	13,796 ms	—	—	1	—	—
Prefill	128K	1	Valid	3,523.0	—	36,301 ms	—	—	1	—	—
Sustained decode	0	1	Valid	66.7	66.7	126 ms	15.0 ms	1.0	0	2,002	30.0 s
Sustained decode	0	2	Valid	121.3	60.7	162 ms	16.5 ms	2.0	0	3,636	30.0 s
Sustained decode	0	4	Valid	207.8	52.1	226 ms	19.2 ms	4.0	0	6,234	30.0 s
Sustained decode	0	8	Valid	319.8	40.7	324 ms	24.5 ms	8.0	0	9,592	30.0 s
Sustained decode	0	16	Valid	495.2	31.9	504 ms	31.4 ms	16.0	0	14,848	30.0 s
Sustained decode	0	32	Valid	748.5	24.3	738 ms	41.1 ms	32.0	0	22,458	30.0 s
Sustained decode	0	64	Valid	1,110.9	18.4	1,117 ms	54.3 ms	64.0	0	33,316	30.0 s
Sustained decode	0	128	Valid	1,613.3	12.9	1,515 ms	77.7 ms	128.0	0	48,384	30.0 s
Sustained decode	16K	1	Valid	65.0	65.0	619 ms	15.4 ms	1.0	0	1,949	30.0 s
Sustained decode	16K	2	Valid	111.4	55.4	854 ms	18.1 ms	2.0	0	3,340	30.0 s
Sustained decode	16K	4	Valid	198.5	48.9	1,269 ms	20.5 ms	4.0	0	5,954	30.0 s
Sustained decode	16K	8	Valid	332.7	40.1	2,079 ms	24.9 ms	8.0	0	9,978	30.0 s
Sustained decode	16K	16	Valid	499.8	29.0	3,679 ms	34.5 ms	16.0	0	14,992	30.0 s
Sustained decode	16K	32	Valid	696.2	18.7	7,069 ms	53.6 ms	32.0	0	20,880	30.0 s
Sustained decode	16K	64	Valid	948.6	10.9	14,094 ms	91.8 ms	64.0	0	28,448	30.0 s
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	63.3	63.4	510 ms	15.8 ms	1.0	0	1,897	30.0 s
Sustained decode	32K	2	Valid	108.6	54.2	684 ms	18.5 ms	2.0	0	3,257	30.0 s
Sustained decode	32K	4	Valid	192.5	47.7	944 ms	21.0 ms	4.0	0	5,776	30.0 s
Sustained decode	32K	8	Valid	315.5	38.6	1,518 ms	25.9 ms	8.0	0	9,464	30.0 s
Sustained decode	32K	16	Valid	465.3	27.7	2,643 ms	36.1 ms	16.0	0	13,952	30.0 s
Sustained decode	32K	32	Valid	632.3	17.7	5,039 ms	56.6 ms	32.0	0	18,960	30.0 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	60.9	61.0	1,026 ms	16.4 ms	1.0	0	1,825	30.0 s
Sustained decode	64K	2	Valid	104.5	52.0	1,321 ms	19.2 ms	2.0	0	3,136	30.0 s
Sustained decode	64K	4	Valid	181.3	44.4	1,872 ms	22.5 ms	4.0	0	5,436	30.0 s
Sustained decode	64K	8	Valid	288.4	34.3	2,953 ms	29.1 ms	8.0	0	8,648	30.0 s
Sustained decode	64K	16	Valid	408.0	22.9	5,187 ms	43.6 ms	16.0	0	12,240	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	56.9	57.1	932 ms	17.5 ms	1.0	0	1,705	30.0 s
Sustained decode	128K	2	Valid	97.8	49.0	1,014 ms	20.4 ms	2.0	0	2,934	30.0 s
Sustained decode	128K	4	Valid	161.5	40.3	1,550 ms	24.8 ms	4.0	0	4,844	30.0 s
Sustained decode	128K	8	Valid	244.5	30.2	2,394 ms	33.1 ms	8.0	0	7,336	30.0 s
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	50.6	50.9	2,056 ms	19.6 ms	1.0	0	1,518	30.0 s
Sustained decode	248K	2	Valid	86.8	43.4	2,538 ms	23.1 ms	2.0	0	2,604	30.0 s
Sustained decode	248K	4	Valid	108.5	26.7	3,665 ms	37.5 ms	4.0	0	3,255	30.0 s
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	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.