Speed test matrix

Tess-4-9B on RTX 3090 24GB

vllm · 0.32.0 · BF166 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 31, 2026, 21:38 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: 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 migtissera/Tess-4-9B --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code --language-model-only --kv-cache-dtype int4_per_token_head --attention-backend TRITON_ATTN

Prefill

prompt tok/s by context length

8K context5,8021,412 ms TTFT

16K context5,0053,241 ms TTFT

32K context3,9158,249 ms TTFT

64K context2,71523,727 ms TTFT

128K context1,69475,960 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	48.639 ms TTFT	95.257 ms TTFT	187.490 ms TTFT	368.8144 ms TTFT	684.8223 ms TTFT	1,165.5362 ms TTFT	Over capacity	Over capacity
16K	45.5503 ms TTFT	88.1762 ms TTFT	162.01,269 ms TTFT	296.42,273 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	42.7634 ms TTFT	81.9944 ms TTFT	141.91,545 ms TTFT	246.52,783 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	38.2663 ms TTFT	72.0965 ms TTFT	115.41,560 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	31.61,969 ms TTFT	58.22,913 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	22.0Requests queued3,037 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	5,802.0	—	1,412 ms	—	—	1	—	—
Prefill	16K	1	Valid	5,005.0	—	3,241 ms	—	—	1	—	—
Prefill	32K	1	Valid	3,915.0	—	8,249 ms	—	—	1	—	—
Prefill	64K	1	Valid	2,715.0	—	23,727 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,694.0	—	75,960 ms	—	—	1	—	—
Sustained decode	0	1	Valid	48.6	48.7	39 ms	20.5 ms	1.0	0	1,459	30.0 s
Sustained decode	0	2	Valid	95.2	47.6	57 ms	21.0 ms	2.0	0	2,856	30.0 s
Sustained decode	0	4	Valid	187.4	46.9	90 ms	21.3 ms	4.0	0	5,620	30.0 s
Sustained decode	0	8	Valid	368.8	46.2	144 ms	21.6 ms	8.0	0	11,064	30.0 s
Sustained decode	0	16	Valid	684.8	42.9	223 ms	23.3 ms	16.0	0	20,544	30.0 s
Sustained decode	0	32	Valid	1,165.5	36.5	362 ms	27.4 ms	32.0	0	34,944	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	45.5	45.5	503 ms	22.0 ms	1.0	0	1,364	30.0 s
Sustained decode	16K	2	Valid	88.1	43.7	762 ms	22.9 ms	2.0	0	2,640	30.0 s
Sustained decode	16K	4	Valid	162.0	39.7	1,269 ms	25.2 ms	4.0	0	4,856	30.0 s
Sustained decode	16K	8	Valid	296.4	35.4	2,273 ms	28.3 ms	8.0	0	8,888	30.0 s
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	42.7	42.7	634 ms	23.4 ms	1.0	0	1,280	30.0 s
Sustained decode	32K	2	Valid	81.9	40.6	944 ms	24.6 ms	2.0	0	2,456	30.0 s
Sustained decode	32K	4	Valid	141.9	34.6	1,545 ms	28.9 ms	4.0	0	4,256	30.0 s
Sustained decode	32K	8	Valid	246.5	29.1	2,783 ms	34.3 ms	8.0	0	7,392	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	38.2	38.2	663 ms	26.2 ms	1.0	0	1,146	30.0 s
Sustained decode	64K	2	Valid	72.0	35.7	965 ms	28.0 ms	2.0	0	2,160	30.0 s
Sustained decode	64K	4	Valid	115.4	28.2	1,560 ms	35.5 ms	4.0	0	3,460	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	31.6	31.6	1,969 ms	31.6 ms	1.0	0	948	30.0 s
Sustained decode	128K	2	Valid	58.2	28.4	2,913 ms	35.2 ms	2.0	0	1,746	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	Requests queued	22.0	24.0	3,037 ms	41.6 ms	0.9	0	660	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.