Speed test matrix

Hy-MT2-7B-FP8 on RTX 3090 24GB

vllm · 0.32.0 · FP86 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, 06:20 UTC

Hardware

Type: Discrete GPU
GPU: RTX 3090
VRAM: 24 GB
CPU: Intel Core i7-9700K (8C)
RAM: 32 GB
OS: Ubuntu 24.04.5

Engine

Engine: vllm
Version: 0.32.0
Quantization: FP8
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), 420 W limit, PCIe Gen3 x16 | Intel Core i7-9700K 8C | 32 GB DDR4 | Ubuntu 24.04.5, NVIDIA driver 580.178.04
vllm serve tencent/Hy-MT2-7B-FP8 --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --enforce-eager --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context3,9972,062 ms TTFT

16K context2,8915,650 ms TTFT

32K context1,84917,594 ms TTFT

64K context1,06361,073 ms TTFT

128K context568.0228,366 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	19.84,050 ms TTFT	38.74,094 ms TTFT	77.6163 ms TTFT	155.1169 ms TTFT	305.4197 ms TTFT	632.9202 ms TTFT	Over capacity	Over capacity
16K	19.8140 ms TTFT	39.2178 ms TTFT	78.1222 ms TTFT	155.1298 ms TTFT	295.4421 ms TTFT	Over capacity	Over capacity	Over capacity
32K	19.8209 ms TTFT	38.9259 ms TTFT	78.2357 ms TTFT	141.0468 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	19.8372 ms TTFT	38.5472 ms TTFT	65.7649 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	19.2Requests queued658 ms TTFT	31.3Requests queued846 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	Over capacity	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	3,997.0	—	2,062 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,891.0	—	5,650 ms	—	—	1	—	—
Prefill	32K	1	Valid	1,849.0	—	17,594 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,063.0	—	61,073 ms	—	—	1	—	—
Prefill	128K	1	Valid	568.0	—	228,366 ms	—	—	1	—	—
Sustained decode	0	1	Valid	19.8	19.8	4,050 ms	50.5 ms	1.0	0	594	30.0 s
Sustained decode	0	2	Valid	38.7	19.3	4,094 ms	51.7 ms	2.0	0	1,160	30.0 s
Sustained decode	0	4	Valid	77.6	19.4	163 ms	51.7 ms	4.0	0	2,324	30.0 s
Sustained decode	0	8	Valid	155.1	19.4	169 ms	51.6 ms	8.0	0	4,648	30.0 s
Sustained decode	0	16	Valid	305.4	19.1	197 ms	52.5 ms	16.0	0	9,152	30.0 s
Sustained decode	0	32	Valid	632.9	19.8	202 ms	50.6 ms	32.0	0	18,976	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	19.8	19.8	140 ms	50.6 ms	1.0	0	593	30.0 s
Sustained decode	16K	2	Valid	39.2	19.6	178 ms	51.0 ms	2.0	0	1,174	30.0 s
Sustained decode	16K	4	Valid	78.1	19.5	222 ms	51.2 ms	4.0	0	2,344	30.0 s
Sustained decode	16K	8	Valid	155.1	19.4	298 ms	51.6 ms	8.0	0	4,648	30.0 s
Sustained decode	16K	16	Valid	295.4	18.5	421 ms	54.1 ms	16.0	0	8,848	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	19.8	19.8	209 ms	50.5 ms	1.0	0	593	30.0 s
Sustained decode	32K	2	Valid	38.9	19.5	259 ms	51.4 ms	2.0	0	1,166	30.0 s
Sustained decode	32K	4	Valid	78.2	19.5	357 ms	51.2 ms	4.0	0	2,344	30.0 s
Sustained decode	32K	8	Valid	141.0	17.6	468 ms	56.8 ms	8.0	0	4,224	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	19.8	19.8	372 ms	50.6 ms	1.0	0	592	30.0 s
Sustained decode	64K	2	Valid	38.5	19.2	472 ms	52.0 ms	2.0	0	1,154	30.0 s
Sustained decode	64K	4	Valid	65.7	16.4	649 ms	60.9 ms	4.0	0	1,972	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	Requests queued	19.2	19.3	658 ms	51.9 ms	1.0	0	576	30.0 s
Sustained decode	128K	2	Requests queued	31.3	15.8	846 ms	63.3 ms	1.9	0	938	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	Over capacity	—	—	—	—	0.6	0	0	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	—	—	—	—	—	—	—	—

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.