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 29, 2026, 12:51 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: 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 (3090 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 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 context3,4022,423 ms TTFT

16K context2,3736,877 ms TTFT

32K context1,49321,743 ms TTFT

64K context853.075,961 ms TTFT

120K context486.0249,844 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	99.636 ms TTFT	193.920 ms TTFT	354.529 ms TTFT	627.135 ms TTFT	1,026.864 ms TTFT	1,640.083 ms TTFT	Over capacity	Over capacity
16K	64.897 ms TTFT	116.0140 ms TTFT	162.2205 ms TTFT	219.3304 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	47.5212 ms TTFT	81.8264 ms TTFT	104.2341 ms TTFT	130.5494 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	30.9348 ms TTFT	51.1495 ms TTFT	60.2673 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	19.1Requests queued663 ms TTFT	30.3Requests queued894 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	3,402.0	—	2,423 ms	—	—	1	—	—
Prefill	16K	1	Valid	2,373.0	—	6,877 ms	—	—	1	—	—
Prefill	32K	1	Valid	1,493.0	—	21,743 ms	—	—	1	—	—
Prefill	64K	1	Valid	853.0	—	75,961 ms	—	—	1	—	—
Prefill	120K	1	Valid	486.0	—	249,844 ms	—	—	1	—	—
Sustained decode	0	1	Valid	99.6	100.8	36 ms	9.9 ms	1.0	0	2,989	30.0 s
Sustained decode	0	2	Valid	193.9	98.3	20 ms	10.2 ms	2.0	0	5,817	30.0 s
Sustained decode	0	4	Valid	354.5	90.7	29 ms	11.0 ms	4.0	0	10,634	30.0 s
Sustained decode	0	8	Valid	627.1	81.2	35 ms	12.3 ms	8.0	0	18,812	30.0 s
Sustained decode	0	16	Valid	1,026.8	67.5	64 ms	14.8 ms	16.0	0	30,816	30.0 s
Sustained decode	0	32	Valid	1,640.0	55.4	83 ms	18.1 ms	32.0	0	49,184	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	64.8	65.0	97 ms	15.4 ms	1.0	0	1,944	30.0 s
Sustained decode	16K	2	Valid	116.0	58.2	140 ms	17.2 ms	2.0	0	3,480	30.0 s
Sustained decode	16K	4	Valid	162.2	40.7	205 ms	24.6 ms	4.0	0	4,864	30.0 s
Sustained decode	16K	8	Valid	219.3	27.5	304 ms	36.4 ms	8.0	0	6,576	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	47.5	47.6	212 ms	21.0 ms	1.0	0	1,425	30.0 s
Sustained decode	32K	2	Valid	81.8	41.0	264 ms	24.4 ms	2.0	0	2,455	30.0 s
Sustained decode	32K	4	Valid	104.2	26.1	341 ms	38.4 ms	4.0	0	3,124	30.0 s
Sustained decode	32K	8	Valid	130.5	16.3	494 ms	61.2 ms	8.0	0	3,912	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	30.9	30.9	348 ms	32.4 ms	1.0	0	925	30.0 s
Sustained decode	64K	2	Valid	51.1	25.6	495 ms	39.1 ms	2.0	0	1,532	30.0 s
Sustained decode	64K	4	Valid	60.2	15.0	673 ms	66.4 ms	4.0	0	1,804	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	19.1	19.1	663 ms	52.2 ms	1.0	0	572	30.0 s
Sustained decode	120K	2	Requests queued	30.3	15.3	894 ms	65.3 ms	1.9	0	908	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.