【瞎折腾/mi50 32G/ubuntu】mi50显卡ubuntu运行大模型开坑(三)安装风扇并且控制转速

o0o_-_2025-05-12 22:44

目录

说在前面

  • 主板：未知x99
  • cpu：E5 2666v3
  • 显卡：Mi 50 32G
  • 系统：ubuntu 22.04
  • 风扇：4pin pwm 外壳风扇
  • 其他：mi50跑模型的时候，散热不好真不行，很容易到90+度

风扇安装

  • 这是拆下来的原装外壳，将上下共六颗螺丝拆下；左侧支架三颗螺丝拆下（不拆的话会卡着，外壳卸不下）

  • 这是装好的外壳

  • 将风扇接口连接到主板的cpu fan上

确认传感器

  • 安装fancontrol

    shell 复制代码 
    sudo apt install fancontrol

    这个时候如果直接使用pwmconfig，只能看到3个设备

    shell 复制代码 
    lele@lele-X99M-Gaming:~$ sudo pwmconfig
# pwmconfig version 3.6.0
This program will search your sensors for pulse width modulation (pwm)
controls, and test each one to see if it controls a fan on
your motherboard. Note that many motherboards do not have pwm
circuitry installed, even if your sensor chip supports pwm.

We will attempt to briefly stop each fan using the pwm controls.
The program will attempt to restore each fan to full speed
after testing. However, it is ** very important ** that you
physically verify that the fans have been to full speed
after the program has completed.

Found the following devices:
   hwmon0 is nvme
   hwmon1 is coretemp
   hwmon2 is amdgpu

Found the following PWM controls:
   hwmon2/pwm1           current value: 255

Giving the fans some time to reach full speed...
Found the following fan sensors:
   hwmon2/fan1_input     current speed: 0 ... skipping!

There are no working fan sensors, all readings are 0.
Make sure you have a 3-wire fan connected.
You may also need to increase the fan divisors.
See doc/fan-divisors for more information.

    虽然明确看到hwmon2 is amdgpu，但是mi 50是没有风扇的，所以出现了提示There are no working fan sensors, all readings are 0.我还以为hwmon2就是装的外置风扇，结果怎么调都没反应。

  • 接下来我们要手动检测这个外置风扇

确认主板IO型号

  • 确认 X99 主板的 Super I/O 芯片型号

    shell 复制代码 
    sudo apt install lm-sensors
sudo sensors-detect

    执行后，在所有输入的地方输入YES

    shell 复制代码 
    # sensors-detect version 3.6.0
# System: YiChenYu X99M-Gaming [Default string]
# Kernel: 6.8.0-58-generic x86_64
# Processor: Intel(R) Xeon(R) CPU E5-2666 v3 @ 2.90GHz (6/63/2)

This program will help you determine which kernel modules you need
to load to use lm_sensors most effectively. It is generally safe
and recommended to accept the default answers to all questions,
unless you know what you're doing.

Some south bridges, CPUs or memory controllers contain embedded sensors.
Do you want to scan for them? This is totally safe. (YES/no): YES

    最终可以看到型号nct6775：

    shell 复制代码 
    Now follows a summary of the probes I have just done.
Just press ENTER to continue:

Driver `lm63':
  * Bus `SMBus I801 adapter at f000'
    Busdriver `i2c_i801', I2C address 0x4c
    Chip `National Semiconductor LM96163' (confidence: 6)

Driver `nct6775':
  * ISA bus, address 0xa20
    Chip `Nuvoton NCT5532D/NCT6779D Super IO Sensors' (confidence: 9)

Driver `coretemp':
  * Chip `Intel digital thermal sensor' (confidence: 9)

To load everything that is needed, add this to /etc/modules:
#----cut here----
# Chip drivers
coretemp
lm63
nct6775
#----cut here----

加载内核模块

  • 使用lsmod确认模块是否加载

    shell 复制代码 
    lsmod | grep nct6775

    如果没有输出，则加载对应模块：

    shell 复制代码 
    sudo modprobe nct6775

    确认加载完成：

    shell 复制代码 
    lele@lele-X99M-Gaming:~$ lsmod | grep nct6775
nct6775                40960  0
nct6775_core          106496  1 nct6775
hwmon_vid              12288  1 nct6775

再次验证传感器

  • 这个时候查看/sys/class/hwmon目录，发现多了一个设备：

    shell 复制代码 
    lele@lele-X99M-Gaming:~$ ls /sys/class/hwmon
hwmon0  hwmon1  hwmon2  hwmon3

  • 查看hwmon3内容，发现有多个pwm

    shell 复制代码 
    lele@lele-X99M-Gaming:~$ ls /sys/class/hwmon/hwmon3/ | grep pwm|grep -v _
pwm1
pwm2
pwm3
pwm4
pwm5

  • 所以我们需要确认哪一个是gpu风扇

    shell 复制代码 
    sudo pwmconfig
# pwmconfig version 3.6.0
This program will search your sensors for pulse width modulation (pwm)
controls, and test each one to see if it controls a fan on
your motherboard. Note that many motherboards do not have pwm
circuitry installed, even if your sensor chip supports pwm.

We will attempt to briefly stop each fan using the pwm controls.
The program will attempt to restore each fan to full speed
after testing. However, it is ** very important ** that you
physically verify that the fans have been to full speed
after the program has completed.

Found the following devices:
   hwmon0 is nvme
   hwmon1 is coretemp
   hwmon2 is amdgpu
   hwmon3 is nct6779

Found the following PWM controls:
   hwmon2/pwm1           current value: 255
   hwmon3/pwm1           current value: 255
   hwmon3/pwm2           current value: 56
hwmon3/pwm2 is currently setup for automatic speed control.
In general, automatic mode is preferred over manual mode, as
it is more efficient and it reacts faster. Are you sure that
you want to setup this output for manual control? (n)
   hwmon3/pwm3           current value: 255
   hwmon3/pwm4           current value: 255
   hwmon3/pwm5           current value: 255

Giving the fans some time to reach full speed...
Found the following fan sensors:
   hwmon2/fan1_input     current speed: 0 ... skipping!
   hwmon3/fan1_input     current speed: 1646 RPM
   hwmon3/fan2_input     current speed: 1239 RPM
   hwmon3/fan3_input     current speed: 0 ... skipping!
   hwmon3/fan4_input     current speed: 0 ... skipping!
   hwmon3/fan5_input     current speed: 0 ... skipping!

Warning!!! This program will stop your fans, one at a time,
for approximately 5 seconds each!!!
This may cause your processor temperature to rise!!!
If you do not want to do this hit control-C now!!!
Hit return to continue:

Testing pwm control hwmon2/pwm1 ...
  hwmon3/fan1_input ... speed was 1646 now 1652
    no correlation
  hwmon3/fan2_input ... speed was 1239 now 1239
    no correlation

No correlations were detected.
There is either no fan connected to the output of hwmon2/pwm1,
or the connected fan has no rpm-signal connected to one of
the tested fan sensors. (Note: not all motherboards have
the pwm outputs connected to the fan connectors,
check out the hardware database on http://www.almico.com/forumindex.php)

Did you see/hear a fan stopping during the above test (n)? n

Testing pwm control hwmon3/pwm1 ...
  hwmon3/fan1_input ... speed was 1646 now 983
    It appears that fan hwmon3/fan1_input
    is controlled by pwm hwmon3/pwm1
Would you like to generate a detailed correlation (y)? y
    PWM 255 FAN 1668
    PWM 0 FAN 1017

  hwmon3/fan2_input ... speed was 1239 now 1241
    no correlation

Testing pwm control hwmon3/pwm3 ...
  hwmon3/fan1_input ... speed was 1646 now 1702
    no correlation
  hwmon3/fan2_input ... speed was 1239 now 1419
    no correlation

No correlations were detected.
There is either no fan connected to the output of hwmon3/pwm3,
or the connected fan has no rpm-signal connected to one of
the tested fan sensors. (Note: not all motherboards have
the pwm outputs connected to the fan connectors,
check out the hardware database on http://www.almico.com/forumindex.php)

Did you see/hear a fan stopping during the above test (n)?

Testing pwm control hwmon3/pwm4 ...
  hwmon3/fan1_input ... speed was 1646 now 1677
    no correlation
  hwmon3/fan2_input ... speed was 1239 now 1421
    no correlation

No correlations were detected.
There is either no fan connected to the output of hwmon3/pwm4,
or the connected fan has no rpm-signal connected to one of
the tested fan sensors. (Note: not all motherboards have
the pwm outputs connected to the fan connectors,
check out the hardware database on http://www.almico.com/forumindex.php)

Did you see/hear a fan stopping during the above test (n)? n

Testing pwm control hwmon3/pwm5 ...
  hwmon3/fan1_input ... speed was 1646 now 1677
    no correlation
  hwmon3/fan2_input ... speed was 1239 now 1288
    no correlation

No correlations were detected.
There is either no fan connected to the output of hwmon3/pwm5,
or the connected fan has no rpm-signal connected to one of
the tested fan sensors. (Note: not all motherboards have
the pwm outputs connected to the fan connectors,
check out the hardware database on http://www.almico.com/forumindex.php)

Did you see/hear a fan stopping during the above test (n)? n

Testing is complete.
Please verify that all fans have returned to their normal speed.

    再次执行pwmconfig，这个时候程序会开始调试机器上所有pwm风扇，它会先把风扇转速拉满，然后慢慢降低，这样你就知道对应关系了，这里，我知道了hwmon3/pwm2是gpu风扇

手动控制风扇

  • 启用手动控制

    shell 复制代码 
    echo 1 | sudo tee /sys/class/hwmon/hwmon3/pwm2_enable

  • 转速拉满(0最低，255最高)

    shell 复制代码 
    echo 255 | sudo tee /sys/class/hwmon/hwmon3/pwm2

    这样就可以看到风扇呼呼转了，声音贼大

其他

  • 自动控制转速这里就没继续研究了，如果需要的话应该要将gpu温度和fancontrol关联起来，可以问问deepseek怎么操作

  • 测试在满速状态下，跑模型温度大概在75度左右

    shell 复制代码 
    ========================================== ROCm System Management Interface ==========================================
==================================================== Concise Info ====================================================
Device  [Model : Revision]    Temp    Power     Partitions      SCLK     MCLK     Fan     Perf  PwrCap  VRAM%  GPU%
        Name (20 chars)       (Edge)  (Socket)  (Mem, Compute)
======================================================================================================================
0       [0x0834 : 0x01]       75.0°C  164.0W    N/A, N/A        1485Mhz  1000Mhz  100.0%  auto  225.0W   63%   100%
        Radeon Instinct MI50
======================================================================================================================
================================================ End of ROCm SMI Log =================================================
相关推荐
鹿鸣天涯8 小时前
使用VMware Workstation Pro搭建Ubuntu服务器虚拟机
ubuntu
知星小度S8 小时前
系统核心解析：深入操作系统内部机制——进程管理与控制指南（一）【进程/PCB】
linux·运维·服务器·进程
linjoe9912 小时前
【Deep Learning】Ubuntu配置深度学习环境
人工智能·深度学习·ubuntu
码农101号13 小时前
运维安全05 - iptables规则保存与恢复
运维·网络·安全
Empty_77713 小时前
SELinux安全上下文
linux·服务器·安全
bug攻城狮14 小时前
解决Ubuntu中apt-get -y安装时弹出交互提示的问题
linux·运维·ubuntu
xiachong2714 小时前
ubuntu18.04安装PCL1.14
linux·ubuntu
夜阑珊夭夭14 小时前
linux自定义网卡名字
linux·运维
czhc114007566314 小时前
Linux912 shell:$# $1 $?；RHEL 8 AppStream BaseOS
linux
佛天华15 小时前
centos 时间校准
linux·运维·centos
热门推荐
01KGG转MP3工具|非KGM文件|解密音频02UV安装并设置国内源03GitHub 镜像站点04conda中设置镜像地址（附所有可换的地址）05A股预测还能更准？开源大模型Kronos带你跑通预测+回测全流程0646个Nano-banana 精选提示词，持续更新中07UV 工具安装与国内镜像源配置指南08突破百度网盘的下载限速，两种方法教会你【超详细】09智能库存管理的需求预测模型：从业务痛点到落地代码的完整实践10教你如何认证 Gemini 教育优惠的二次验证，薅个 1年的 Gemini Pro 会员