Speed test matrix

Qwen3.8-27B-INT4 on RTX 3090 24GB

vllm · 0.32.0 · INT4 W4A16 g1285 contexts (0–128K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 21:27 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: INT4 W4A16 g128
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: none. KV cache: int4_per_token_head
# 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 RedHatAI/Qwen3.8-27B-INT4 --max-model-len 139264 --gpu-memory-utilization 0.94 --max-num-seqs 54 --enable-prefix-caching --trust-remote-code --language-model-only --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN --max-cudagraph-capture-size 16

Prefill

prompt tok/s by context length

8K context1,5005,489 ms TTFT

16K context1,26912,817 ms TTFT

32K context978.033,049 ms TTFT

64K context669.096,317 ms TTFT

128K context411.0313,290 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	45.589 ms TTFT	84.4290 ms TTFT	162.1295 ms TTFT	312.7552 ms TTFT	391.8Requests queued854 ms TTFT	Over capacity	Over capacity	Over capacity
16K	40.11,082 ms TTFT	71.21,652 ms TTFT	126.42,722 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	36.02,852 ms TTFT	62.24,315 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	29.91,633 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	19.4Requests queued4,686 ms TTFT	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	1,500.0	—	5,489 ms	—	—	1	—	—
Prefill	16K	1	Valid	1,269.0	—	12,817 ms	—	—	1	—	—
Prefill	32K	1	Valid	978.0	—	33,049 ms	—	—	1	—	—
Prefill	64K	1	Valid	669.0	—	96,317 ms	—	—	1	—	—
Prefill	128K	1	Valid	411.0	—	313,290 ms	—	—	1	—	—
Sustained decode	0	1	Valid	45.5	45.6	89 ms	21.9 ms	1.0	0	1,365	30.0 s
Sustained decode	0	2	Valid	84.4	42.3	290 ms	23.7 ms	2.0	0	2,530	30.0 s
Sustained decode	0	4	Valid	162.1	40.6	295 ms	24.6 ms	4.0	0	4,860	30.0 s
Sustained decode	0	8	Valid	312.7	39.2	552 ms	25.5 ms	8.0	0	9,376	30.0 s
Sustained decode	0	16	Requests queued	391.8	33.6	854 ms	29.8 ms	12.4	0	11,308	28.9 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	40.1	40.1	1,082 ms	24.9 ms	1.0	0	1,201	30.0 s
Sustained decode	16K	2	Valid	71.2	35.1	1,652 ms	28.5 ms	2.0	0	2,136	30.0 s
Sustained decode	16K	4	Valid	126.4	30.2	2,722 ms	33.1 ms	4.0	0	3,792	30.0 s
Sustained decode	16K	8	Over capacity	—	—	—	—	—	—	—	—
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	Valid	36.0	36.0	2,852 ms	27.8 ms	1.0	0	1,079	30.0 s
Sustained decode	32K	2	Valid	62.2	29.9	4,315 ms	33.5 ms	2.0	0	1,864	30.0 s
Sustained decode	32K	4	Over capacity	—	—	—	—	—	—	—	—
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	Valid	29.9	29.9	1,633 ms	33.4 ms	1.0	0	897	30.0 s
Sustained decode	64K	2	Over capacity	—	—	—	—	—	—	—	—
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	128K	1	Requests queued	19.4	22.5	4,686 ms	44.5 ms	0.8	0	580	30.0 s
Sustained decode	128K	2	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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.