Speed test matrix

Nemotron-Cascade-2-30B-A3B-AWQ-4bit on RTX 3090 24GB

vllm · 0.32.0 · AWQ 4-bit6 contexts (0–176K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Oct 30, 2026, 22:05 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: AWQ 4-bit
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 cyankiwi/Nemotron-Cascade-2-30B-A3B-AWQ-4bit --max-model-len 188416 --gpu-memory-utilization 0.94 --max-num-seqs 104 --enable-prefix-caching --trust-remote-code

Prefill

prompt tok/s by context length

8K context9,139920 ms TTFT

16K context8,9701,855 ms TTFT

32K context8,1894,042 ms TTFT

64K context7,0439,371 ms TTFT

128K context5,51123,922 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	150.133 ms TTFT	267.3185 ms TTFT	434.572 ms TTFT	679.4113 ms TTFT	974.8Requests queued157 ms TTFT	Over capacity	Over capacity	Over capacity
16K	146.0251 ms TTFT	259.0366 ms TTFT	420.1602 ms TTFT	646.11,089 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	143.0268 ms TTFT	251.4393 ms TTFT	407.9620 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
64K	138.6285 ms TTFT	238.8392 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	130.2626 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
176K	124.8577 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	9,139.0	—	920 ms	—	—	1	—	—
Prefill	16K	1	Valid	8,970.0	—	1,855 ms	—	—	1	—	—
Prefill	32K	1	Valid	8,189.0	—	4,042 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,043.0	—	9,371 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,511.0	—	23,922 ms	—	—	1	—	—
Sustained decode	0	1	Valid	150.1	150.2	33 ms	6.7 ms	1.0	0	4,503	30.0 s
Sustained decode	0	2	Valid	267.3	133.8	185 ms	7.5 ms	2.0	0	8,018	30.0 s
Sustained decode	0	4	Valid	434.5	109.0	72 ms	9.2 ms	4.0	0	13,032	30.0 s
Sustained decode	0	8	Valid	679.4	85.3	113 ms	11.7 ms	8.0	0	20,376	30.0 s
Sustained decode	0	16	Requests queued	974.8	61.8	157 ms	16.2 ms	15.9	0	29,239	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	146.0	146.1	251 ms	6.8 ms	1.0	0	4,380	30.0 s
Sustained decode	16K	2	Valid	259.0	129.1	366 ms	7.7 ms	2.0	0	7,768	30.0 s
Sustained decode	16K	4	Valid	420.1	104.1	602 ms	9.6 ms	4.0	0	12,600	30.0 s
Sustained decode	16K	8	Valid	646.1	79.1	1,089 ms	12.6 ms	8.0	0	19,376	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	143.0	143.2	268 ms	7.0 ms	1.0	0	4,291	30.0 s
Sustained decode	32K	2	Valid	251.4	125.4	393 ms	8.0 ms	2.0	0	7,540	30.0 s
Sustained decode	32K	4	Valid	407.9	101.0	620 ms	9.9 ms	4.0	0	12,236	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	138.6	138.7	285 ms	7.2 ms	1.0	0	4,156	30.0 s
Sustained decode	64K	2	Valid	238.8	119.1	392 ms	8.4 ms	2.0	0	7,163	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	128K	1	Valid	130.2	130.4	626 ms	7.7 ms	1.0	0	3,906	30.0 s
Sustained decode	128K	2	Over capacity	—	—	—	—	—	—	—	—
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	176K	1	Valid	124.8	124.9	577 ms	8.0 ms	1.0	0	3,743	30.0 s
Sustained decode	176K	2	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	4	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	176K	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.