[GH-ISSUE #1836] 推理引擎空闲状态下CPU负载过高 #1146

Open
opened 2026-10-29 14:49:14 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @45min-on-the-git on GitHub (Feb 6, 2026).
Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/1836

Originally assigned to: @ovowei on GitHub.

Reminder

  • I have read the above rules and searched the existing issues.

System Info

CPU: 2 * AMD EPYC 9275F 24-Core Processor
RAM: 16 * DDR5 4800 48G
GPU: 1 * 5090
NVIDIA Driver Version: 570.195.03
CUDA Version: 12.8

KTransformers CLI v0.1.0

Python 3.12.12
平台 Linux 6.11.3-061103-generic
CUDA 12.8

Packages:

kt-kernel 0.5.1
sglang 0.0.0.dev8805+g384f824c3.d20260124 (PyPI)

Reproduction

使用如下命令运行推理引擎时,即使再推理引擎没有进行任何推理引擎时,CPU也有两个核心占用为100%,导致CPU在待机情况下温度略高,请问是否有什么降低待机时CPU占用率的方法?

命令:

uv run kt run /models/MiniMax-M2.1/ \
--host 0.0.0.0 \
--port 9810 \
--served-model-name DeepSeek-V3.1 \
--max-running-requests 4 \
--tool-call-parser minimax-m2 \
--reasoning-parser minimax \
--kt-method FP8 \
--kt-cpuinfer 40 \
--kt-threadpool-count 2 \
--kt-num-gpu-experts 8 \
--chunked-prefill-size 12288 \
--max-total-tokens 50000 \
--kt-gpu-prefill-token-threshold 2048 \
--log-level debug


# top页面

top - 15:34:13 up 3 days, 19:48,  4 users,  load average: 2.27, 2.24, 2.20
Threads: 6270 total,   4 running, 6241 sleeping,   0 stopped,  25 zombie
%Cpu(s):  2.2 us,  0.2 sy,  0.0 ni, 97.7 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem : 773526.9 total,  64471.8 free, 258718.7 used, 455698.2 buff/cache
MiB Swap:   8192.0 total,   7756.5 free,    435.5 used. 514808.2 avail Mem

    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
  13007 root      20   0  587.5g 423.6g 209.4g R  99.9  56.1     91,43 sglang::schedul
  13543 root      20   0  587.5g 423.6g 209.4g R  99.9  56.1     91,40 sglang::schedul
   4387 root      20   0   16.4g   1.2g 248416 S   4.3   0.2 210:33.80 python3
3992828 root      20   0   20372  12404   3188 R   2.0   0.0   0:01.04 top
1485686 ubuntu    20   0  609432 179888  28648 S   1.3   0.0  65:44.74 python
  13516 root      20   0  587.5g 423.6g 209.4g S   1.0  56.1 104:11.12 numa_1_t_9
   4389 root      20   0 7064600 886640 145596 S   0.7   0.1  23:34.47 python3
  13486 root      20   0  587.5g 423.6g 209.4g S   0.7  56.1 104:17.51 numa_0_t_4
  13487 root      20   0  587.5g 423.6g 209.4g S   0.7  56.1 103:31.76 numa_0_t_5

Others

No response

Originally created by @45min-on-the-git on GitHub (Feb 6, 2026). Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/1836 Originally assigned to: @ovowei on GitHub. ### Reminder - [x] I have read the above rules and searched the existing issues. ### System Info CPU: 2 * AMD EPYC 9275F 24-Core Processor RAM: 16 * DDR5 4800 48G GPU: 1 * 5090 NVIDIA Driver Version: 570.195.03 CUDA Version: 12.8 KTransformers CLI v0.1.0 Python 3.12.12 平台 Linux 6.11.3-061103-generic CUDA 12.8 Packages: kt-kernel 0.5.1 sglang 0.0.0.dev8805+g384f824c3.d20260124 (PyPI) ### Reproduction ```text 使用如下命令运行推理引擎时,即使再推理引擎没有进行任何推理引擎时,CPU也有两个核心占用为100%,导致CPU在待机情况下温度略高,请问是否有什么降低待机时CPU占用率的方法? 命令: uv run kt run /models/MiniMax-M2.1/ \ --host 0.0.0.0 \ --port 9810 \ --served-model-name DeepSeek-V3.1 \ --max-running-requests 4 \ --tool-call-parser minimax-m2 \ --reasoning-parser minimax \ --kt-method FP8 \ --kt-cpuinfer 40 \ --kt-threadpool-count 2 \ --kt-num-gpu-experts 8 \ --chunked-prefill-size 12288 \ --max-total-tokens 50000 \ --kt-gpu-prefill-token-threshold 2048 \ --log-level debug # top页面 top - 15:34:13 up 3 days, 19:48, 4 users, load average: 2.27, 2.24, 2.20 Threads: 6270 total, 4 running, 6241 sleeping, 0 stopped, 25 zombie %Cpu(s): 2.2 us, 0.2 sy, 0.0 ni, 97.7 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st MiB Mem : 773526.9 total, 64471.8 free, 258718.7 used, 455698.2 buff/cache MiB Swap: 8192.0 total, 7756.5 free, 435.5 used. 514808.2 avail Mem PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 13007 root 20 0 587.5g 423.6g 209.4g R 99.9 56.1 91,43 sglang::schedul 13543 root 20 0 587.5g 423.6g 209.4g R 99.9 56.1 91,40 sglang::schedul 4387 root 20 0 16.4g 1.2g 248416 S 4.3 0.2 210:33.80 python3 3992828 root 20 0 20372 12404 3188 R 2.0 0.0 0:01.04 top 1485686 ubuntu 20 0 609432 179888 28648 S 1.3 0.0 65:44.74 python 13516 root 20 0 587.5g 423.6g 209.4g S 1.0 56.1 104:11.12 numa_1_t_9 4389 root 20 0 7064600 886640 145596 S 0.7 0.1 23:34.47 python3 13486 root 20 0 587.5g 423.6g 209.4g S 0.7 56.1 104:17.51 numa_0_t_4 13487 root 20 0 587.5g 423.6g 209.4g S 0.7 56.1 103:31.76 numa_0_t_5 ``` ### Others _No response_
Author
Owner

@ErvinXie commented on GitHub (Feb 28, 2026):

这是一个已知问题,与 #1858 相同。kt-kernel 的 TaskQueue worker 线程在空闲时使用了 busy-wait 循环,会导致 CPU 核心占满。我们已确认问题并正在排查修复中。


This reply was co-authored by an AI agent and a human.

<!-- gh-comment-id:3976696882 --> @ErvinXie commented on GitHub (Feb 28, 2026): 这是一个已知问题,与 #1858 相同。kt-kernel 的 TaskQueue worker 线程在空闲时使用了 busy-wait 循环,会导致 CPU 核心占满。我们已确认问题并正在排查修复中。 --- *This reply was co-authored by an AI agent and a human.*
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
kvcache-ai/ktransformers#1146
No description provided.