Speed test matrix

Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8 on H200 NVL 141GB

vllm · 0.26.1rc1.dev638+g52be12cfa · FP86 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 4, 2026, 19:57 UTC

Hardware

Type: Discrete GPU
GPU: H200 NVL
VRAM: 141 GB
OS: linux

Engine

Engine: vllm
Version: 0.26.1rc1.dev638+g52be12cfa
Quantization: FP8
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2
Commit: 52be12cfac0c

Run

Served model: nemotron
Revision: main
Benchmark tool: llm-inference-bench v0.4.29
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 4,096 tokens

Launch command

"$HOME/env_h200-nvidia-nemotron/bin/vllm" serve nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-FP8 --served-model-name nemotron --trust-remote-code --tensor-parallel-size 1 --reasoning-parser nemotron_v3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 262144 --gpu-memory-utilization 0.886040 --host 0.0.0.0 --port 8000 --api-key <redacted> --video-pruning-rate 0.5 --max-num-seqs 384 --allowed-local-media-path / --media-io-kwargs '{"video":{"fps":2,"num_frames":256}}' --kv-cache-dtype fp8

Prefill

prompt tok/s by context length

8K context59.2K138 ms TTFT

16K context57.3K283 ms TTFT

32K context59K547 ms TTFT

64K context55.3K1,164 ms TTFT

128K context43K2,988 ms TTFT

251K context29.8K8,457 ms TTFT

256K context29.2K8,793 ms TTFT

Sustained decode

total tok/s · rows are context length, columns are simultaneous requests

Context	1	2	4	8	16	32	64	128
8K	385.480 ms TTFT	756.1127 ms TTFT	1,207.6163 ms TTFT	2,153.1247 ms TTFT	2,424.5352 ms TTFT	4,789.5550 ms TTFT	1,210.8988 ms TTFT	2,334.51,751 ms TTFT
16K	401.5109 ms TTFT	632.0157 ms TTFT	1,026.5171 ms TTFT	2,051.7245 ms TTFT	2,393.0335 ms TTFT	4,419.8579 ms TTFT	1,383.5934 ms TTFT	3,643.41,510 ms TTFT
32K	348.2105 ms TTFT	750.7144 ms TTFT	1,069.1193 ms TTFT	1,207.1247 ms TTFT	2,470.5253 ms TTFT	4,245.3347 ms TTFT	2,085.91,180 ms TTFT	3,139.22,179 ms TTFT
64K	339.9172 ms TTFT	606.5232 ms TTFT	1,026.8240 ms TTFT	1,295.6263 ms TTFT	2,287.0516 ms TTFT	2,059.51,178 ms TTFT	2,645.42,201 ms TTFT	3,816.54,365 ms TTFT
128K	324.3295 ms TTFT	575.0347 ms TTFT	931.6409 ms TTFT	1,291.5387 ms TTFT	1,790.21,155 ms TTFT	2,224.42,441 ms TTFT	3,030.24,953 ms TTFT	3,584.79,589 ms TTFT
251K	301.9555 ms TTFT	524.5602 ms TTFT	754.2566 ms TTFT	1,122.6624 ms TTFT	1,456.32,785 ms TTFT	1,592.95,168 ms TTFT	2,116.89,734 ms TTFT	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	59.2K	—	138 ms	—	—	1	—	—
Prefill	16K	1	Valid	57.3K	—	283 ms	—	—	1	—	—
Prefill	32K	1	Valid	59K	—	547 ms	—	—	1	—	—
Prefill	64K	1	Valid	55.3K	—	1,164 ms	—	—	1	—	—
Prefill	128K	1	Valid	43K	—	2,988 ms	—	—	1	—	—
Prefill	251K	1	Valid	29.8K	—	8,457 ms	—	—	1	—	—
Prefill	256K	1	Valid	29.2K	—	8,793 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	385.4	2270.5	80 ms	0.4 ms	1.0	2	7,812	20.3 s
Sustained decode	8K	2	Valid	756.1	3825.0	127 ms	0.3 ms	2.0	4	17,312	22.9 s
Sustained decode	8K	4	Valid	1,207.6	687.7	163 ms	2.2 ms	4.0	8	31,685	26.2 s
Sustained decode	8K	8	Valid	2,153.1	1560.2	247 ms	0.7 ms	8.0	16	64,591	30.0 s
Sustained decode	8K	16	Valid	2,424.5	199.0	352 ms	5.0 ms	15.6	16	72,728	30.0 s
Sustained decode	8K	32	Valid	4,789.5	250.0	550 ms	5.1 ms	31.1	32	143,688	30.0 s
Sustained decode	8K	64	Valid	1,210.8	91.8	988 ms	10.9 ms	64.0	0	36,318	30.0 s
Sustained decode	8K	128	Valid	2,334.5	63.6	1,751 ms	15.7 ms	128.0	0	70,016	30.0 s
Sustained decode	16K	1	Valid	401.5	1198.0	109 ms	0.8 ms	1.0	2	8,313	20.7 s
Sustained decode	16K	2	Valid	632.0	1597.2	157 ms	0.7 ms	2.0	4	18,959	30.0 s
Sustained decode	16K	4	Valid	1,026.5	264.6	171 ms	3.8 ms	4.0	8	30,795	30.0 s
Sustained decode	16K	8	Valid	2,051.7	918.5	245 ms	1.3 ms	7.7	16	61,553	30.0 s
Sustained decode	16K	16	Valid	2,393.0	180.7	335 ms	5.5 ms	16.0	16	71,778	30.0 s
Sustained decode	16K	32	Valid	4,419.8	134.0	579 ms	7.5 ms	32.0	32	132,593	30.0 s
Sustained decode	16K	64	Valid	1,383.5	88.3	934 ms	11.3 ms	64.0	0	41,504	30.0 s
Sustained decode	16K	128	Valid	3,643.4	61.7	1,510 ms	16.2 ms	128.0	0	109,315	30.0 s
Sustained decode	32K	1	Valid	348.2	352.2	105 ms	2.8 ms	1.0	2	7,492	21.5 s
Sustained decode	32K	2	Valid	750.7	1063.6	144 ms	1.8 ms	2.0	4	16,174	21.5 s
Sustained decode	32K	4	Valid	1,069.1	276.9	193 ms	3.6 ms	4.0	8	32,073	30.0 s
Sustained decode	32K	8	Valid	1,207.1	212.9	247 ms	4.7 ms	8.0	8	36,213	30.0 s
Sustained decode	32K	16	Valid	2,470.5	164.1	253 ms	6.1 ms	15.9	16	74,115	30.0 s
Sustained decode	32K	32	Valid	4,245.3	123.2	347 ms	8.2 ms	31.7	32	127,362	30.0 s
Sustained decode	32K	64	Valid	2,085.9	80.3	1,180 ms	12.5 ms	64.0	0	62,570	30.0 s
Sustained decode	32K	128	Valid	3,139.2	55.1	2,179 ms	18.2 ms	128.0	0	94,208	30.0 s
Sustained decode	64K	1	Valid	339.9	344.7	172 ms	2.9 ms	1.0	2	10,196	30.0 s
Sustained decode	64K	2	Valid	606.5	312.1	232 ms	3.2 ms	1.9	4	16,810	27.7 s
Sustained decode	64K	4	Valid	1,026.8	251.5	240 ms	4.0 ms	4.0	8	30,802	30.0 s
Sustained decode	64K	8	Valid	1,295.6	204.4	263 ms	4.9 ms	7.9	8	38,869	30.0 s
Sustained decode	64K	16	Valid	2,287.0	148.5	516 ms	6.7 ms	15.8	16	68,604	30.0 s
Sustained decode	64K	32	Valid	2,059.5	104.5	1,178 ms	9.6 ms	32.0	0	61,774	30.0 s
Sustained decode	64K	64	Valid	2,645.4	69.0	2,201 ms	14.5 ms	64.0	0	79,367	30.0 s
Sustained decode	64K	128	Valid	3,816.5	44.4	4,365 ms	22.5 ms	128.0	0	114,521	30.0 s
Sustained decode	128K	1	Valid	324.3	332.2	295 ms	3.0 ms	1.0	2	9,727	30.0 s
Sustained decode	128K	2	Valid	575.0	294.8	347 ms	3.4 ms	2.0	4	17,252	30.0 s
Sustained decode	128K	4	Valid	931.6	231.5	409 ms	4.3 ms	4.0	4	27,947	30.0 s
Sustained decode	128K	8	Valid	1,291.5	172.5	387 ms	5.8 ms	7.9	8	38,747	30.0 s
Sustained decode	128K	16	Valid	1,790.2	119.5	1,155 ms	8.4 ms	16.0	6	53,681	30.0 s
Sustained decode	128K	32	Valid	2,224.4	81.7	2,441 ms	12.2 ms	32.0	0	66,727	30.0 s
Sustained decode	128K	64	Valid	3,030.2	52.2	4,953 ms	19.2 ms	64.0	0	90,890	30.0 s
Sustained decode	128K	128	Valid	3,584.7	32.2	9,589 ms	31.1 ms	128.0	0	107,491	30.0 s
Sustained decode	251K	1	Valid	301.9	309.1	555 ms	3.2 ms	1.0	2	9,058	30.0 s
Sustained decode	251K	2	Valid	524.5	274.2	602 ms	3.6 ms	1.9	4	15,736	30.0 s
Sustained decode	251K	4	Valid	754.2	210.0	566 ms	4.8 ms	3.9	4	22,624	30.0 s
Sustained decode	251K	8	Valid	1,122.6	150.0	624 ms	6.7 ms	7.8	8	33,679	30.0 s
Sustained decode	251K	16	Valid	1,456.3	96.7	2,785 ms	10.3 ms	16.0	0	43,680	30.0 s
Sustained decode	251K	32	Valid	1,592.9	61.0	5,168 ms	16.4 ms	32.0	0	47,783	30.0 s
Sustained decode	251K	64	Valid	2,116.8	36.3	9,734 ms	27.6 ms	64.0	0	63,499	30.0 s
Sustained decode	251K	128	Over capacity	—	—	—	—	—	—	—	—

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