Speed test matrix

gemma-4-12B-it-qat-w4a16-ct on RTX 3090 24GB

vllm · 0.32.0 · W4A166 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 29, 2026, 17:12 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: W4A16
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 google/gemma-4-12B-it-qat-w4a16-ct --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 context3,0122,660 ms TTFT

16K context2,6426,009 ms TTFT

32K context2,17014,559 ms TTFT

64K context1,60639,246 ms TTFT

128K context1,056119,243 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	68.345 ms TTFT	129.550 ms TTFT	235.083 ms TTFT	400.988 ms TTFT	620.1118 ms TTFT	949.5187 ms TTFT	1,397.0341 ms TTFT	Over capacity
16K	49.498 ms TTFT	Looped	Looped	Looped	Looped	Over capacity	Over capacity	Over capacity
32K	43.2227 ms TTFT	Looped	160.8344 ms TTFT	Looped	Over capacity	Over capacity	Over capacity	Over capacity
64K	37.4346 ms TTFT	74.2467 ms TTFT	136.9684 ms TTFT	Looped	Over capacity	Over capacity	Over capacity	Over capacity
128K	29.5784 ms TTFT	58.9968 ms TTFT	104.91,372 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	20.6Requests queued1,561 ms TTFT	40.4Requests queued2,026 ms TTFT	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. Looped: the model kept repeating the same text, so the benchmark discarded this measurement. 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	3,012.0	—	2,660 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,642.0	—	6,009 ms	—	—	1	—	—
Prefill	32K	1	Valid	2,170.0	—	14,559 ms	—	—	1	—	—
Prefill	64K	1	Valid	1,606.0	—	39,246 ms	—	—	1	—	—
Prefill	128K	1	Valid	1,056.0	—	119,243 ms	—	—	1	—	—
Sustained decode	0	1	Valid	68.3	68.8	45 ms	14.5 ms	1.0	0	2,048	30.0 s
Sustained decode	0	2	Valid	129.5	65.6	50 ms	15.2 ms	2.0	0	3,884	30.0 s
Sustained decode	0	4	Valid	235.0	60.1	83 ms	16.7 ms	4.0	0	7,048	30.0 s
Sustained decode	0	8	Valid	400.9	51.9	88 ms	19.3 ms	8.0	0	12,024	30.0 s
Sustained decode	0	16	Valid	620.1	41.1	118 ms	24.3 ms	16.0	0	18,592	30.0 s
Sustained decode	0	32	Valid	949.5	32.4	187 ms	30.9 ms	32.0	0	28,480	30.0 s
Sustained decode	0	64	Valid	1,397.0	23.6	341 ms	42.3 ms	64.0	0	41,920	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	49.4	49.8	98 ms	20.1 ms	1.0	0	1,482	30.0 s
Sustained decode	16K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	4	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	8	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	16	Looped	—	—	—	—	—	0	0	—
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	43.2	43.2	227 ms	23.2 ms	1.0	0	1,294	30.0 s
Sustained decode	32K	2	Looped	—	—	—	—	—	0	0	—
Sustained decode	32K	4	Valid	160.8	40.2	344 ms	24.9 ms	4.0	0	4,820	30.0 s
Sustained decode	32K	8	Looped	—	—	—	—	—	0	0	—
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	37.4	37.5	346 ms	26.6 ms	1.0	0	1,123	30.0 s
Sustained decode	64K	2	Valid	74.2	37.1	467 ms	26.9 ms	2.0	0	2,224	30.0 s
Sustained decode	64K	4	Valid	136.9	34.2	684 ms	29.2 ms	4.0	0	4,104	30.0 s
Sustained decode	64K	8	Looped	—	—	—	—	—	0	0	—
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	29.5	29.6	784 ms	33.8 ms	1.0	0	885	30.0 s
Sustained decode	128K	2	Valid	58.9	29.5	968 ms	33.9 ms	2.0	0	1,767	30.0 s
Sustained decode	128K	4	Valid	104.9	26.1	1,372 ms	38.2 ms	4.0	0	3,148	30.0 s
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	20.6	21.4	1,561 ms	46.8 ms	1.0	0	619	30.0 s
Sustained decode	248K	2	Requests queued	40.4	21.3	2,026 ms	47.0 ms	1.9	0	1,212	30.0 s
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	—	—	—	—	—	—	—	—