Speed test matrix

granite-4.2-3b-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, 19:56 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
# 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-4.2-3b-nvfp4 --max-model-len 131072 --gpu-memory-utilization 0.94 --max-num-seqs 128 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context10.8K763 ms TTFT

16K context8,9721,819 ms TTFT

32K context6,4245,054 ms TTFT

64K context4,13515,665 ms TTFT

120K context2,57547,102 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	199.718 ms TTFT	362.720 ms TTFT	653.016 ms TTFT	1,084.820 ms TTFT	1,719.645 ms TTFT	Over capacity	Over capacity	Over capacity
16K	151.740 ms TTFT	245.354 ms TTFT	397.489 ms TTFT	502.3134 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	124.271 ms TTFT	183.490 ms TTFT	275.9141 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	90.6135 ms TTFT	121.6181 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
120K	61.3262 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity

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	10.8K	—	763 ms	—	—	1	—	—
Prefill	16K	1	Valid	8,972.0	—	1,819 ms	—	—	1	—	—
Prefill	32K	1	Valid	6,424.0	—	5,054 ms	—	—	1	—	—
Prefill	64K	1	Valid	4,135.0	—	15,665 ms	—	—	1	—	—
Prefill	120K	1	Valid	2,575.0	—	47,102 ms	—	—	1	—	—
Sustained decode	0	1	Valid	199.7	202.8	18 ms	4.9 ms	1.0	0	5,990	30.0 s
Sustained decode	0	2	Valid	362.7	186.4	20 ms	5.4 ms	2.0	0	10,880	30.0 s
Sustained decode	0	4	Valid	653.0	169.8	16 ms	5.9 ms	4.0	0	19,587	30.0 s
Sustained decode	0	8	Valid	1,084.8	144.0	20 ms	6.9 ms	8.0	0	32,536	30.0 s
Sustained decode	0	16	Valid	1,719.6	116.3	45 ms	8.6 ms	16.0	0	51,584	30.0 s
Sustained decode	0	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	151.7	152.5	40 ms	6.6 ms	1.0	0	4,550	30.0 s
Sustained decode	16K	2	Valid	245.3	123.4	54 ms	8.1 ms	2.0	0	7,358	30.0 s
Sustained decode	16K	4	Valid	397.4	100.4	89 ms	10.0 ms	4.0	0	11,920	30.0 s
Sustained decode	16K	8	Valid	502.3	63.8	134 ms	15.7 ms	8.0	0	15,064	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	124.2	124.5	71 ms	8.0 ms	1.0	0	3,727	30.0 s
Sustained decode	32K	2	Valid	183.4	91.9	90 ms	10.9 ms	2.0	0	5,502	30.0 s
Sustained decode	32K	4	Valid	275.9	69.3	141 ms	14.4 ms	4.0	0	8,276	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	90.6	90.8	135 ms	11.0 ms	1.0	0	2,718	30.0 s
Sustained decode	64K	2	Valid	121.6	60.8	181 ms	16.4 ms	2.0	0	3,646	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	120K	1	Valid	61.3	61.4	262 ms	16.3 ms	1.0	0	1,838	30.0 s
Sustained decode	120K	2	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

Over capacity: not run because this many requests at this context would exceed the server's KV cache.