Speed test matrix

Tess-4-9B on RTX 3090 24GB

vllm · 0.32.0 · BF166 contexts (0–168K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 23:18 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: BF16
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. 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 migtissera/Tess-4-9B --max-model-len 180224 --gpu-memory-utilization 0.94 --max-num-seqs 104 --enable-prefix-caching --trust-remote-code --language-model-only --speculative-config {"method":"mtp","num_speculative_tokens":2} --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context5,1811,581 ms TTFT

16K context4,3453,733 ms TTFT

32K context3,3059,770 ms TTFT

64K context2,24128,746 ms TTFT

128K context1,39692,189 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	62.538 ms TTFT	119.865 ms TTFT	244.798 ms TTFT	439.4156 ms TTFT	389.2Requests queued178 ms TTFT	Over capacity	Over capacity	Over capacity
16K	43.21,061 ms TTFT	84.51,902 ms TTFT	165.82,890 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
32K	34.91,443 ms TTFT	63.02,602 ms TTFT	119.74,309 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	21.21,844 ms TTFT	40.73,316 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	11.43,581 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
168K	8.65,851 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	5,181.0	—	1,581 ms	—	—	1	—	—
Prefill	16K	1	Valid	4,345.0	—	3,733 ms	—	—	1	—	—
Prefill	32K	1	Valid	3,305.0	—	9,770 ms	—	—	1	—	—
Prefill	64K	1	Valid	2,241.0	—	28,746 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,396.0	—	92,189 ms	—	—	1	—	—
Sustained decode	0	1	Valid	62.5	63.7	38 ms	15.7 ms	1.0	0	1,874	30.0 s
Sustained decode	0	2	Valid	119.8	60.6	65 ms	16.5 ms	2.0	0	3,592	30.0 s
Sustained decode	0	4	Valid	244.7	62.4	98 ms	16.0 ms	4.0	0	7,340	30.0 s
Sustained decode	0	8	Valid	439.4	55.5	156 ms	18.0 ms	8.0	0	13,182	30.0 s
Sustained decode	0	16	Requests queued	389.2	46.1	178 ms	21.7 ms	9.0	0	11,650	29.9 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	43.2	43.3	1,061 ms	23.1 ms	1.0	0	1,297	30.0 s
Sustained decode	16K	2	Valid	84.5	41.9	1,902 ms	23.9 ms	2.0	0	2,535	30.0 s
Sustained decode	16K	4	Valid	165.8	39.7	2,890 ms	25.2 ms	4.0	0	4,971	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	34.9	35.0	1,443 ms	28.6 ms	1.0	0	1,046	29.9 s
Sustained decode	32K	2	Valid	63.0	31.4	2,602 ms	31.9 ms	2.0	0	1,887	30.0 s
Sustained decode	32K	4	Valid	119.7	29.2	4,309 ms	34.3 ms	4.0	0	3,587	30.0 s
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	21.2	21.4	1,844 ms	46.7 ms	1.0	0	636	30.0 s
Sustained decode	64K	2	Valid	40.7	20.7	3,316 ms	48.4 ms	2.0	0	1,221	30.0 s
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	Valid	11.4	11.6	3,581 ms	86.3 ms	1.0	0	341	29.9 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	—	—	—	—	—	—	—	—
Sustained decode	168K	1	Valid	8.6	8.8	5,851 ms	114.1 ms	1.0	0	259	30.0 s
Sustained decode	168K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	168K	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.