Speed test matrix

granite-4.2-8b-nvfp4 on RTX 3090 24GB

vllm · 0.32.0 · NVFP45 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, 11:16 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: NVFP4
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 ibm-granite/granite-4.2-8b-nvfp4 --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 context2,9352,807 ms TTFT

16K context2,1007,771 ms TTFT

32K context1,31924,627 ms TTFT

64K context755.085,796 ms TTFT

120K context430.0282,233 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	97.334 ms TTFT	188.521 ms TTFT	342.227 ms TTFT	601.230 ms TTFT	975.745 ms TTFT	1,557.459 ms TTFT	Over capacity	Over capacity
16K	62.486 ms TTFT	109.9123 ms TTFT	152.5199 ms TTFT	202.7282 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	45.8152 ms TTFT	76.9218 ms TTFT	96.8326 ms TTFT	119.6430 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	29.7291 ms TTFT	47.6418 ms TTFT	55.6549 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	18.4Requests queued553 ms TTFT	28.2Requests queued768 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. 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	2,935.0	—	2,807 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,100.0	—	7,771 ms	—	—	1	—	—
Prefill	32K	1	Valid	1,319.0	—	24,627 ms	—	—	1	—	—
Prefill	64K	1	Valid	755.0	—	85,796 ms	—	—	1	—	—
Prefill	120K	1	Valid	430.0	—	282,233 ms	—	—	1	—	—
Sustained decode	0	1	Valid	97.3	98.4	34 ms	10.2 ms	1.0	0	2,918	30.0 s
Sustained decode	0	2	Valid	188.5	95.6	21 ms	10.5 ms	2.0	0	5,654	30.0 s
Sustained decode	0	4	Valid	342.2	87.6	27 ms	11.4 ms	4.0	0	10,264	30.0 s
Sustained decode	0	8	Valid	601.2	77.9	30 ms	12.8 ms	8.0	0	18,032	30.0 s
Sustained decode	0	16	Valid	975.7	64.2	45 ms	15.6 ms	16.0	0	29,264	30.0 s
Sustained decode	0	32	Valid	1,557.4	52.5	59 ms	19.1 ms	32.0	0	46,688	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	62.4	62.5	86 ms	16.0 ms	1.0	0	1,871	30.0 s
Sustained decode	16K	2	Valid	109.9	55.1	123 ms	18.1 ms	2.0	0	3,296	30.0 s
Sustained decode	16K	4	Valid	152.5	38.3	199 ms	26.1 ms	4.0	0	4,572	30.0 s
Sustained decode	16K	8	Valid	202.7	25.5	282 ms	39.3 ms	8.0	0	6,080	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	45.8	45.8	152 ms	21.8 ms	1.0	0	1,373	30.0 s
Sustained decode	32K	2	Valid	76.9	38.5	218 ms	26.0 ms	2.0	0	2,306	30.0 s
Sustained decode	32K	4	Valid	96.8	24.3	326 ms	41.2 ms	4.0	0	2,904	30.0 s
Sustained decode	32K	8	Valid	119.6	15.0	430 ms	66.8 ms	8.0	0	3,584	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	29.7	29.7	291 ms	33.6 ms	1.0	0	891	30.0 s
Sustained decode	64K	2	Valid	47.6	23.8	418 ms	42.0 ms	2.0	0	1,426	30.0 s
Sustained decode	64K	4	Valid	55.6	13.9	549 ms	71.9 ms	4.0	0	1,664	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	120K	1	Requests queued	18.4	18.4	553 ms	54.5 ms	1.0	0	550	30.0 s
Sustained decode	120K	2	Requests queued	28.2	14.2	768 ms	70.5 ms	1.9	0	844	30.0 s
Sustained decode	120K	4	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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.