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 Oct 31, 2026, 05:51 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: 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 cyankiwi/Nemotron-Cascade-2-30B-A3B-AWQ-4bit --max-model-len 262144 --gpu-memory-utilization 0.94 --max-num-seqs 105 --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 context8,1691,030 ms TTFT

16K context6,8862,416 ms TTFT

32K context5,2056,359 ms TTFT

64K context3,52518,727 ms TTFT

128K context2,14461,490 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	145.033 ms TTFT	257.750 ms TTFT	422.876 ms TTFT	667.2116 ms TTFT	993.5160 ms TTFT	1,075.9Requests queued245 ms TTFT	1,073.4Requests queued267 ms TTFT	Over capacity
16K	131.1217 ms TTFT	230.2318 ms TTFT	390.7523 ms TTFT	616.1943 ms TTFT	918.91,792 ms TTFT	Over capacity	Over capacity	Over capacity
32K	120.0501 ms TTFT	211.8738 ms TTFT	350.71,208 ms TTFT	541.72,168 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
64K	102.51,501 ms TTFT	180.42,266 ms TTFT	306.93,678 ms TTFT	520.26,621 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
128K	79.35,069 ms TTFT	140.47,557 ms TTFT	230.812,511 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity	Over capacity
248K	45.1Requests queued6,153 ms TTFT	61.2Requests queued9,063 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	8,169.0	—	1,030 ms	—	—	1	—	—
Prefill	16K	1	Valid	6,886.0	—	2,416 ms	—	—	1	—	—
Prefill	32K	1	Valid	5,205.0	—	6,359 ms	—	—	1	—	—
Prefill	64K	1	Valid	3,525.0	—	18,727 ms	—	—	1	—	—
Prefill	128K	1	Valid	2,144.0	—	61,490 ms	—	—	1	—	—
Sustained decode	0	1	Valid	145.0	145.4	33 ms	6.9 ms	1.0	0	4,349	30.0 s
Sustained decode	0	2	Valid	257.7	129.1	50 ms	7.7 ms	2.0	0	7,730	30.0 s
Sustained decode	0	4	Valid	422.8	106.1	76 ms	9.4 ms	4.0	0	12,684	30.0 s
Sustained decode	0	8	Valid	667.2	83.7	116 ms	11.9 ms	8.0	0	20,016	30.0 s
Sustained decode	0	16	Valid	993.5	62.5	160 ms	16.0 ms	16.0	0	29,792	30.0 s
Sustained decode	0	32	Requests queued	1,075.9	52.4	245 ms	19.1 ms	21.0	0	32,277	30.0 s
Sustained decode	0	64	Requests queued	1,073.4	52.2	267 ms	19.2 ms	21.0	0	32,193	30.0 s
Sustained decode	0	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	131.1	131.3	217 ms	7.6 ms	1.0	0	3,932	30.0 s
Sustained decode	16K	2	Valid	230.2	114.9	318 ms	8.7 ms	2.0	0	6,905	30.0 s
Sustained decode	16K	4	Valid	390.7	97.1	523 ms	10.3 ms	4.0	0	11,720	30.0 s
Sustained decode	16K	8	Valid	616.1	75.7	943 ms	13.2 ms	8.0	0	18,480	30.0 s
Sustained decode	16K	16	Valid	918.9	55.2	1,792 ms	18.1 ms	16.0	0	27,552	30.0 s
Sustained decode	16K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	120.0	120.1	501 ms	8.3 ms	1.0	0	3,598	30.0 s
Sustained decode	32K	2	Valid	211.8	105.3	738 ms	9.5 ms	2.0	0	6,354	30.0 s
Sustained decode	32K	4	Valid	350.7	86.1	1,208 ms	11.6 ms	4.0	0	10,521	30.0 s
Sustained decode	32K	8	Valid	541.7	64.7	2,168 ms	15.4 ms	8.0	0	16,248	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	102.5	102.7	1,501 ms	9.7 ms	1.0	0	3,076	30.0 s
Sustained decode	64K	2	Valid	180.4	88.4	2,266 ms	11.3 ms	2.0	0	5,410	30.0 s
Sustained decode	64K	4	Valid	306.9	72.2	3,678 ms	13.8 ms	4.0	0	9,204	30.0 s
Sustained decode	64K	8	Valid	520.2	56.7	6,621 ms	17.6 ms	8.0	0	15,600	30.0 s
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	79.3	79.4	5,069 ms	12.6 ms	1.0	0	2,379	30.0 s
Sustained decode	128K	2	Valid	140.4	65.8	7,557 ms	15.3 ms	2.0	0	4,210	30.0 s
Sustained decode	128K	4	Valid	230.8	47.6	12,511 ms	21.1 ms	4.0	0	6,924	30.0 s
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	Requests queued	45.1	55.5	6,153 ms	18.0 ms	0.9	0	1,352	30.0 s
Sustained decode	248K	2	Requests queued	61.2	43.6	9,063 ms	23.4 ms	1.7	0	1,836	30.0 s
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.