Speed test matrix

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

vllm · 0.32.0 · AWQ 4-bit6 contexts (0–248K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K

by @Sandbergllm-inference-bench v0.8.0Measured Nov 1, 2026, 00:57 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. KV cache: fp8
# 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 262144 --gpu-memory-utilization 0.94 --max-num-seqs 88 --enable-prefix-caching --trust-remote-code --kv-cache-dtype fp8

Prefill

prompt tok/s by context length

8K context9,394895 ms TTFT

16K context9,2611,797 ms TTFT

32K context8,5793,858 ms TTFT

64K context7,3998,920 ms TTFT

128K context5,81322,681 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	151.433 ms TTFT	269.254 ms TTFT	439.073 ms TTFT	684.6113 ms TTFT	1,020.5158 ms TTFT	953.4Requests queued243 ms TTFT	Over capacity	Over capacity
16K	149.3486 ms TTFT	264.9812 ms TTFT	430.21,381 ms TTFT	667.92,292 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
32K	147.8540 ms TTFT	262.4784 ms TTFT	424.11,286 ms TTFT	652.72,289 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	145.2652 ms TTFT	258.9947 ms TTFT	412.71,518 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
128K	140.5750 ms TTFT	249.61,043 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	132.2641 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	9,394.0	—	895 ms	—	—	1	—	—
Prefill	16K	1	Valid	9,261.0	—	1,797 ms	—	—	1	—	—
Prefill	32K	1	Valid	8,579.0	—	3,858 ms	—	—	1	—	—
Prefill	64K	1	Valid	7,399.0	—	8,920 ms	—	—	1	—	—
Prefill	128K	1	Valid	5,813.0	—	22,681 ms	—	—	1	—	—
Sustained decode	0	1	Valid	151.4	151.5	33 ms	6.6 ms	1.0	0	4,543	30.0 s
Sustained decode	0	2	Valid	269.2	134.7	54 ms	7.4 ms	2.0	0	8,074	30.0 s
Sustained decode	0	4	Valid	439.0	110.0	73 ms	9.1 ms	4.0	0	13,168	30.0 s
Sustained decode	0	8	Valid	684.6	85.9	113 ms	11.6 ms	8.0	0	20,536	30.0 s
Sustained decode	0	16	Valid	1,020.5	64.2	158 ms	15.6 ms	16.0	0	30,608	30.0 s
Sustained decode	0	32	Requests queued	953.4	53.6	243 ms	18.7 ms	18.0	0	28,591	30.0 s
Sustained decode	0	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	149.3	149.4	486 ms	6.7 ms	1.0	0	4,479	30.0 s
Sustained decode	16K	2	Valid	264.9	132.0	812 ms	7.6 ms	2.0	0	7,946	30.0 s
Sustained decode	16K	4	Valid	430.2	106.3	1,381 ms	9.4 ms	4.0	0	12,907	30.0 s
Sustained decode	16K	8	Valid	667.9	80.5	2,292 ms	12.4 ms	8.0	0	20,032	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	147.8	147.8	540 ms	6.8 ms	1.0	0	4,433	30.0 s
Sustained decode	32K	2	Valid	262.4	130.4	784 ms	7.7 ms	2.0	0	7,870	30.0 s
Sustained decode	32K	4	Valid	424.1	103.9	1,286 ms	9.6 ms	4.0	0	12,724	30.0 s
Sustained decode	32K	8	Valid	652.7	77.9	2,289 ms	12.8 ms	8.0	0	19,576	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	145.2	145.3	652 ms	6.9 ms	1.0	0	4,357	30.0 s
Sustained decode	64K	2	Valid	258.9	128.4	947 ms	7.8 ms	2.0	0	7,766	30.0 s
Sustained decode	64K	4	Valid	412.7	100.8	1,518 ms	9.9 ms	4.0	0	12,379	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	128K	1	Valid	140.5	140.6	750 ms	7.1 ms	1.0	0	4,213	30.0 s
Sustained decode	128K	2	Valid	249.6	123.8	1,043 ms	8.1 ms	2.0	0	7,488	30.0 s
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	248K	1	Valid	132.2	132.4	641 ms	7.6 ms	1.0	0	3,966	30.0 s
Sustained decode	248K	2	Over capacity	—	—	—	—	—	—	—	—
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	—	—	—	—	—	—	—	—

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.