Speed test matrix

Kimi-K2.6-NVFP4 on H200 SXM 141GB ×8

vllm · 0.29.0 · NVFP46 contexts (8K–251K) × 8 concurrency levels (1–128) · prefill at 8K, 16K, 32K, 64K, 128K, 251K, 256K

by @keennaylmx dev importMeasured Sep 17, 2026, 04:59 UTC

Hardware

Type: Discrete GPU
GPU: H200 SXM
GPU count: 8
VRAM: 1128 GB (8 × 141 GB)
OS: linux

Engine

Engine: vllm
Version: 0.29.0
Quantization: NVFP4
Backend: cuda
Repository: github.com/vllm-project/vllm
Build: cuda-13.2

Run

Served model: nvidia/Kimi-K2.6-NVFP4
Revision: main
Benchmark tool: llm-inference-bench v0.6.2
Decode test: Fixed duration
Prefill mode: Measured during decode warm-up
Time per measurement: 30 s
Max output: 4,096 tokens

Launch command

env FLASH_ATTENTION_CUTE_DSL_CACHE_ENABLED=1 VLLM_ENGINE_READY_TIMEOUT_S=3600 "$HOME/env_h200-glm53-vllm-v0290/bin/vllm" serve nvidia/Kimi-K2.6-NVFP4 --served-model-name nvidia/Kimi-K2.6-NVFP4 --trust-remote-code --tensor-parallel-size 8 --reasoning-parser kimi_k2 --enable-auto-tool-choice --tool-call-parser kimi_k2 --max-model-len 262144 --gpu-memory-utilization 0.86 --host 0.0.0.0 --port 8000 --api-key <redacted> --enforce-eager --max-num-batched-tokens 16384 --kernel-config.enable_flashinfer_autotune=False

Prefill

prompt tok/s by context length

8K context17.6K466 ms TTFT

16K context18.7K869 ms TTFT

32K context17.5K1,845 ms TTFT

64K context15.4K4,193 ms TTFT

128K context12.4K10,402 ms TTFT

251K context8,92528,284 ms TTFT

256K context8,87829,001 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	8.6176 ms TTFT	19.3442 ms TTFT	41.3393 ms TTFT	82.7399 ms TTFT	165.8549 ms TTFT	326.7710 ms TTFT	630.81,164 ms TTFT	Over capacity
16K	8.6167 ms TTFT	20.7292 ms TTFT	41.4412 ms TTFT	82.9562 ms TTFT	165.1698 ms TTFT	326.21,083 ms TTFT	Over capacity	Over capacity
32K	8.6220 ms TTFT	20.6335 ms TTFT	41.3517 ms TTFT	82.5734 ms TTFT	165.11,101 ms TTFT	326.31,750 ms TTFT	Over capacity	Over capacity
64K	8.7277 ms TTFT	20.6474 ms TTFT	41.0703 ms TTFT	82.61,075 ms TTFT	165.01,704 ms TTFT	Over capacity	Over capacity	Over capacity
128K	8.8448 ms TTFT	20.6694 ms TTFT	41.21,167 ms TTFT	82.51,749 ms TTFT	Over capacity	Over capacity	Over capacity	Over capacity
251K	8.4774 ms TTFT	20.61,207 ms TTFT	41.21,901 ms TTFT	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	17.6K	—	466 ms	—	—	1	—	—
Prefill	16K	1	Valid	18.7K	—	869 ms	—	—	1	—	—
Prefill	32K	1	Valid	17.5K	—	1,845 ms	—	—	1	—	—
Prefill	64K	1	Valid	15.4K	—	4,193 ms	—	—	1	—	—
Prefill	128K	1	Valid	12.4K	—	10,402 ms	—	—	1	—	—
Prefill	251K	1	Valid	8,925.0	—	28,284 ms	—	—	1	—	—
Prefill	256K	1	Valid	8,878.0	—	29,001 ms	—	—	1	—	—
Sustained decode	8K	1	Valid	8.6	8.6	176 ms	115.8 ms	1.0	0	257	30.0 s
Sustained decode	8K	2	Valid	19.3	9.5	442 ms	105.6 ms	2.0	0	578	29.9 s
Sustained decode	8K	4	Valid	41.3	10.3	393 ms	97.1 ms	4.0	0	1,236	29.9 s
Sustained decode	8K	8	Valid	82.7	10.3	399 ms	96.9 ms	8.0	0	2,480	30.0 s
Sustained decode	8K	16	Valid	165.8	10.3	549 ms	96.6 ms	16.0	0	4,960	29.9 s
Sustained decode	8K	32	Valid	326.7	10.2	710 ms	98.4 ms	32.0	0	9,792	30.0 s
Sustained decode	8K	64	Valid	630.8	9.8	1,164 ms	101.9 ms	64.0	0	18,877	29.9 s
Sustained decode	8K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	1	Valid	8.6	8.6	167 ms	116.5 ms	1.0	0	258	29.8 s
Sustained decode	16K	2	Valid	20.7	10.3	292 ms	97.0 ms	2.0	0	620	30.0 s
Sustained decode	16K	4	Valid	41.4	10.3	412 ms	96.9 ms	4.0	0	1,240	30.0 s
Sustained decode	16K	8	Valid	82.9	10.3	562 ms	96.8 ms	8.0	0	2,480	29.9 s
Sustained decode	16K	16	Valid	165.1	10.3	698 ms	97.1 ms	16.0	0	4,944	29.9 s
Sustained decode	16K	32	Valid	326.2	10.1	1,083 ms	98.8 ms	32.0	0	9,759	29.9 s
Sustained decode	16K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	16K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	1	Valid	8.6	8.6	220 ms	116.5 ms	1.0	0	257	30.0 s
Sustained decode	32K	2	Valid	20.6	10.3	335 ms	97.0 ms	2.0	0	618	29.9 s
Sustained decode	32K	4	Valid	41.3	10.3	517 ms	97.1 ms	4.0	0	1,236	29.9 s
Sustained decode	32K	8	Valid	82.5	10.3	734 ms	97.2 ms	8.0	0	2,472	30.0 s
Sustained decode	32K	16	Valid	165.1	10.3	1,101 ms	97.5 ms	16.0	0	4,944	29.9 s
Sustained decode	32K	32	Valid	326.3	10.1	1,750 ms	99.1 ms	32.0	0	9,759	29.9 s
Sustained decode	32K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	32K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	1	Valid	8.7	8.6	277 ms	115.8 ms	1.0	0	260	30.0 s
Sustained decode	64K	2	Valid	20.6	10.3	474 ms	97.0 ms	2.0	0	618	29.9 s
Sustained decode	64K	4	Valid	41.0	10.2	703 ms	97.7 ms	4.0	0	1,228	29.9 s
Sustained decode	64K	8	Valid	82.6	10.3	1,075 ms	97.5 ms	8.0	0	2,472	29.9 s
Sustained decode	64K	16	Valid	165.0	10.2	1,704 ms	98.0 ms	16.0	0	4,944	30.0 s
Sustained decode	64K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	64K	128	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	128K	1	Valid	8.8	8.7	448 ms	114.8 ms	1.0	0	264	30.0 s
Sustained decode	128K	2	Valid	20.6	10.3	694 ms	97.4 ms	2.0	0	618	30.0 s
Sustained decode	128K	4	Valid	41.2	10.3	1,167 ms	97.5 ms	4.0	0	1,232	29.9 s
Sustained decode	128K	8	Valid	82.5	10.2	1,749 ms	97.6 ms	8.0	0	2,471	29.9 s
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	251K	1	Valid	8.4	8.5	774 ms	117.3 ms	1.0	0	252	29.8 s
Sustained decode	251K	2	Valid	20.6	10.3	1,207 ms	97.3 ms	2.0	0	618	30.0 s
Sustained decode	251K	4	Valid	41.2	10.2	1,901 ms	97.7 ms	4.0	0	1,236	30.0 s
Sustained decode	251K	8	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	16	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	32	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	64	Over capacity	—	—	—	—	—	—	—	—
Sustained decode	251K	128	Over capacity	—	—	—	—	—	—	—	—

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