Speed test matrix

gpt-oss-20b on RTX 3090 24GB

vllm · 0.32.0 · MXFP45 contexts (0–120K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 120K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 02:21 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: MXFP4
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 openai/gpt-oss-20b --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 context6,8711,187 ms TTFT

16K context5,0143,211 ms TTFT

32K context3,2899,728 ms TTFT

64K context1,94232,838 ms TTFT

120K context1,141104,687 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	139.746 ms TTFT	235.851 ms TTFT	346.759 ms TTFT	588.865 ms TTFT	825.378 ms TTFT	1,577.9106 ms TTFT	2,708.2176 ms TTFT	Over capacity
16K	119.160 ms TTFT	190.780 ms TTFT	273.7130 ms TTFT	409.5185 ms TTFT	586.0324 ms TTFT	762.5624 ms TTFT	Over capacity	Over capacity
32K	104.488 ms TTFT	81.0148 ms TTFT	209.1215 ms TTFT	281.0348 ms TTFT	365.2602 ms TTFT	Over capacity	Over capacity	Over capacity
64K	83.6132 ms TTFT	129.8233 ms TTFT	169.3305 ms TTFT	217.3578 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
120K	61.8220 ms TTFT	83.9381 ms TTFT	110.8565 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	6,871.0	—	1,187 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,014.0	—	3,211 ms	—	—	1	—	—
Prefill	32K	1	Valid	3,289.0	—	9,728 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,942.0	—	32,838 ms	—	—	1	—	—
Prefill	120K	1	Valid	1,141.0	—	104,687 ms	—	—	1	—	—
Sustained decode	0	1	Valid	139.7	140.7	46 ms	7.1 ms	1.0	0	4,190	30.0 s
Sustained decode	0	2	Valid	235.8	113.9	51 ms	8.8 ms	2.0	0	7,074	30.0 s
Sustained decode	0	4	Valid	346.7	90.0	59 ms	11.1 ms	4.0	0	10,399	30.0 s
Sustained decode	0	8	Valid	588.8	75.5	65 ms	13.3 ms	8.0	0	17,658	30.0 s
Sustained decode	0	16	Valid	825.3	62.2	78 ms	16.1 ms	16.0	0	24,757	30.0 s
Sustained decode	0	32	Valid	1,577.9	54.9	106 ms	18.2 ms	32.0	0	47,326	30.0 s
Sustained decode	0	64	Valid	2,708.2	47.8	176 ms	20.9 ms	64.0	0	81,248	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	119.1	119.5	60 ms	8.4 ms	1.0	0	3,572	30.0 s
Sustained decode	16K	2	Valid	190.7	94.8	80 ms	10.5 ms	2.0	0	5,719	30.0 s
Sustained decode	16K	4	Valid	273.7	71.2	130 ms	14.0 ms	4.0	0	8,208	30.0 s
Sustained decode	16K	8	Valid	409.5	54.5	185 ms	18.3 ms	8.0	0	12,285	30.0 s
Sustained decode	16K	16	Valid	586.0	39.2	324 ms	25.5 ms	16.0	0	17,581	30.0 s
Sustained decode	16K	32	Valid	762.5	29.4	624 ms	34.1 ms	32.0	0	22,864	30.0 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	104.4	104.7	88 ms	9.6 ms	1.0	0	3,133	30.0 s
Sustained decode	32K	2	Valid	81.0	82.8	148 ms	12.1 ms	2.0	0	2,429	30.0 s
Sustained decode	32K	4	Valid	209.1	59.1	215 ms	16.9 ms	4.0	0	6,271	30.0 s
Sustained decode	32K	8	Valid	281.0	43.6	348 ms	23.0 ms	8.0	0	8,426	30.0 s
Sustained decode	32K	16	Valid	365.2	28.5	602 ms	35.1 ms	16.0	0	10,949	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	83.6	83.8	132 ms	11.9 ms	1.0	0	2,496	29.8 s
Sustained decode	64K	2	Valid	129.8	64.7	233 ms	15.5 ms	2.0	0	3,891	30.0 s
Sustained decode	64K	4	Valid	169.3	44.3	305 ms	22.6 ms	4.0	0	5,079	30.0 s
Sustained decode	64K	8	Valid	217.3	29.9	578 ms	33.4 ms	8.0	0	6,516	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	120K	1	Valid	61.8	62.0	220 ms	16.1 ms	1.0	0	1,855	30.0 s
Sustained decode	120K	2	Valid	83.9	46.7	381 ms	21.4 ms	2.0	0	2,515	30.0 s
Sustained decode	120K	4	Valid	110.8	30.4	565 ms	32.9 ms	4.0	0	3,324	30.0 s
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	—	—	—	—	—	—	—	—

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