Speed test matrix

Qwen3.5-4B-AWQ-4bit on RTX 3090 24GB

vllm · 0.32.0 · 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 Oct 30, 2026, 03:50 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3090
VRAM: 24 GB
CPU: AMD Ryzen 9 7900X (12C)
RAM: 32 GB
OS: Ubuntu 24.04.5

Engine

Engine: vllm
Version: 0.32.0
Quantization: AWQ 4-bit
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: vllm/vllm-openai:v0.32.0-x86_64

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
# 1x RTX 3090 24GB (Zagitta/gpu-unlock), 350 W limit (VBIOS max), PCIe Gen4 x16 | AMD Ryzen 9 7900X 12C | 32 GB DDR5 | Ubuntu 24.04.5, NVIDIA driver 595.99.02
vllm serve cyankiwi/Qwen3.5-4B-AWQ-4bit --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2}

Prefill

prompt tok/s by context length

8K context9,922825 ms TTFT

16K context9,2701,750 ms TTFT

32K context8,1783,949 ms TTFT

64K context6,6069,752 ms TTFT

128K context4,69927,388 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	198.725 ms TTFT	380.052 ms TTFT	716.957 ms TTFT	1,117.098 ms TTFT	1,627.6165 ms TTFT	2,080.9232 ms TTFT	Over capacity	Over capacity
16K	121.9140 ms TTFT	239.1220 ms TTFT	453.9406 ms TTFT	581.1698 ms TTFT	671.01,213 ms TTFT	Over capacity	Over capacity	Over capacity
32K	90.1157 ms TTFT	178.3232 ms TTFT	312.3432 ms TTFT	376.5835 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	53.2253 ms TTFT	109.4353 ms TTFT	195.2685 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	32.5457 ms TTFT	65.7619 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	17.4701 ms TTFT	Over capacity	Over capacity	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	9,922.0	—	825 ms	—	—	1	—	—
Prefill	16K	1	Valid	9,270.0	—	1,750 ms	—	—	1	—	—
Prefill	32K	1	Valid	8,178.0	—	3,949 ms	—	—	1	—	—
Prefill	64K	1	Valid	6,606.0	—	9,752 ms	—	—	1	—	—
Prefill	128K	1	Valid	4,699.0	—	27,388 ms	—	—	1	—	—
Sustained decode	0	1	Valid	198.7	203.5	25 ms	4.9 ms	1.0	0	5,960	30.0 s
Sustained decode	0	2	Valid	380.0	194.1	52 ms	5.2 ms	2.0	0	11,396	30.0 s
Sustained decode	0	4	Valid	716.9	185.2	57 ms	5.4 ms	4.0	0	21,501	30.0 s
Sustained decode	0	8	Valid	1,117.0	145.6	98 ms	6.9 ms	8.0	0	33,511	30.0 s
Sustained decode	0	16	Valid	1,627.6	106.8	165 ms	9.4 ms	16.0	0	48,816	30.0 s
Sustained decode	0	32	Valid	2,080.9	67.0	232 ms	14.9 ms	32.0	0	62,389	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	121.9	123.5	140 ms	8.1 ms	1.0	0	3,657	30.0 s
Sustained decode	16K	2	Valid	239.1	120.4	220 ms	8.3 ms	2.0	0	7,169	30.0 s
Sustained decode	16K	4	Valid	453.9	114.1	406 ms	8.8 ms	4.0	0	13,616	30.0 s
Sustained decode	16K	8	Valid	581.1	72.6	698 ms	13.8 ms	8.0	0	17,429	30.0 s
Sustained decode	16K	16	Valid	671.0	42.7	1,213 ms	23.4 ms	16.0	0	20,103	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	90.1	90.8	157 ms	11.0 ms	1.0	0	2,702	30.0 s
Sustained decode	32K	2	Valid	178.3	89.7	232 ms	11.2 ms	2.0	0	5,350	30.0 s
Sustained decode	32K	4	Valid	312.3	79.6	432 ms	12.6 ms	4.0	0	9,368	30.0 s
Sustained decode	32K	8	Valid	376.5	48.3	835 ms	20.7 ms	8.0	0	11,287	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	53.2	54.1	253 ms	18.5 ms	1.0	0	1,595	30.0 s
Sustained decode	64K	2	Valid	109.4	55.5	353 ms	18.0 ms	2.0	0	3,278	30.0 s
Sustained decode	64K	4	Valid	195.2	49.0	685 ms	20.4 ms	4.0	0	5,853	30.0 s
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	128K	1	Valid	32.5	33.5	457 ms	29.8 ms	1.0	0	974	30.0 s
Sustained decode	128K	2	Valid	65.7	33.3	619 ms	30.0 ms	2.0	0	1,971	30.0 s
Sustained decode	128K	4	Over capacity	—	—	—	—	—	—	—	—
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	17.4	18.1	701 ms	55.3 ms	1.0	0	520	30.0 s
Sustained decode	248K	2	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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