在Linux服务器上更换Nvidia驱动

向右看齐 2022-12-26 10:16 202阅读 0赞

> 本文记录在Linux服务器更换Nvidia驱动的流程。

### 需求 ###

> Linux 服务器上的 1080Ti 显卡驱动为387， CUDA 9，比较老旧，需要更换成可以运行pytorch 1.6的环境。

### 确定当前显卡型号\\操作系统版本\\目标环境 ###

*  查看显卡信息，确定自己的显卡型号：

$ nvidia-smi

或

$ lspci  | grep -i vga

> 输出的设备信息并不是我们熟悉的型号，比如我的输出为：

02:00.0 VGA compatible controller: NVIDIA Corporation Device 1b06 (rev a1)
    04:00.0 VGA compatible controller: NVIDIA Corporation Device 1b06 (rev a1)
    09:00.0 VGA compatible controller: Matrox Electronics Systems Ltd. G200eR2 (rev 01)
    83:00.0 VGA compatible controller: NVIDIA Corporation Device 1b06 (rev a1)
    84:00.0 VGA compatible controller: NVIDIA Corporation Device 1b06 (rev a1)

> 其中 `Device 1b06` 即为 1080Ti
> 
> 可以进一步查看左边编号下的显卡信息

$ lspci -v -s 02:00.0
    ->
    02:00.0 VGA compatible controller: NVIDIA Corporation Device 1b06 (rev a1) (prog-if 00 [VGA controller])
    	Subsystem: Gigabyte Technology Co., Ltd Device 376b
    	Flags: bus master, fast devsel, latency 0, IRQ 83
    	Memory at 91000000 (32-bit, non-prefetchable) [size=16M]
    	Memory at 3bfe0000000 (64-bit, prefetchable) [size=256M]
    	Memory at 3bff0000000 (64-bit, prefetchable) [size=32M]
    	I/O ports at 2000 [size=128]
    	[virtual] Expansion ROM at 92080000 [disabled] [size=512K]
    	Capabilities: <access denied>
    	Kernel driver in use: nvidia
    	Kernel modules: nvidiafb, nouveau, nvidia_drm, nvidia

*  查看系统信息

$ lsb_release -a
    ->
    No LSB modules are available.
    Distributor ID:	Ubuntu
    Description:	Ubuntu 16.04.5 LTS
    Release:	16.04
    Codename:	xenial

*  查看系统位数

$ file /bin/ls
    ->
    /bin/ls: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=d0bc0fb9b3f60f72bbad3c5a1d24c9e2a1fde775, stripped

*  确定需求的Cuda版本

> 比如 tensorflow-gpu 的各个版本对Cuda 和 cuDNN 都有自己的要求：
> 
> https://tensorflow.google.cn/install/source\#linux

<table> 
 <thead> 
  <tr> 
   <th align="left">Version</th> 
   <th align="left">Python version</th> 
   <th align="left">Compiler</th> 
   <th align="left">Build tools</th> 
   <th align="left">cuDNN</th> 
   <th align="left">CUDA</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td align="left">tensorflow-2.3.0</td> 
   <td align="left">3.5-3.8</td> 
   <td align="left">GCC 7.3.1</td> 
   <td align="left">Bazel 3.1.0</td> 
   <td align="left">7.6</td> 
   <td align="left">10.1</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow-2.2.0</td> 
   <td align="left">3.5-3.8</td> 
   <td align="left">GCC 7.3.1</td> 
   <td align="left">Bazel 2.0.0</td> 
   <td align="left">7.6</td> 
   <td align="left">10.1</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow-2.1.0</td> 
   <td align="left">2.7, 3.5-3.7</td> 
   <td align="left">GCC 7.3.1</td> 
   <td align="left">Bazel 0.27.1</td> 
   <td align="left">7.6</td> 
   <td align="left">10.1</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow-2.0.0</td> 
   <td align="left">2.7, 3.3-3.7</td> 
   <td align="left">GCC 7.3.1</td> 
   <td align="left">Bazel 0.26.1</td> 
   <td align="left">7.4</td> 
   <td align="left">10.0</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.15.0</td> 
   <td align="left">2.7, 3.3-3.7</td> 
   <td align="left">GCC 7.3.1</td> 
   <td align="left">Bazel 0.26.1</td> 
   <td align="left">7.4</td> 
   <td align="left">10.0</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.14.0</td> 
   <td align="left">2.7, 3.3-3.7</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.24.1</td> 
   <td align="left">7.4</td> 
   <td align="left">10.0</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.13.1</td> 
   <td align="left">2.7, 3.3-3.7</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.19.2</td> 
   <td align="left">7.4</td> 
   <td align="left">10.0</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.12.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.15.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.11.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.15.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.10.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.15.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.9.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.11.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.8.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.10.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.7.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.9.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.6.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.9.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.5.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.8.0</td> 
   <td align="left">7</td> 
   <td align="left">9</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.4.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.5.4</td> 
   <td align="left">6</td> 
   <td align="left">8</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.3.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.4.5</td> 
   <td align="left">6</td> 
   <td align="left">8</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.2.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.4.5</td> 
   <td align="left">5.1</td> 
   <td align="left">8</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.1.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.4.2</td> 
   <td align="left">5.1</td> 
   <td align="left">8</td> 
  </tr> 
  <tr> 
   <td align="left">tensorflow_gpu-1.0.0</td> 
   <td align="left">2.7, 3.3-3.6</td> 
   <td align="left">GCC 4.8</td> 
   <td align="left">Bazel 0.4.2</td> 
   <td align="left">5.1</td> 
   <td align="left">8</td> 
  </tr> 
 </tbody> 
</table>

> pytorch 同样有各个版本对Cuda的要求:
> 
> https://pytorch.org/get-started/previous-versions/

*  我们需要安装pytorch1.6的环境,因此需要10.2 以上的Cuda版本.
 *  确定安装的驱动版本:

> https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html\#cuda-major-component-versions

<table> 
 <thead> 
  <tr> 
   <th>CUDA Toolkit</th> 
   <th>Linux x86_64 Driver Version</th> 
   <th>Windows x86_64 Driver Version</th> 
  </tr> 
 </thead> 
 <tbody> 
  <tr> 
   <td>CUDA 11.1.1 Update 1</td> 
   <td>&gt;=455.32</td> 
   <td>&gt;=456.81</td> 
  </tr> 
  <tr> 
   <td>CUDA 11.1 GA</td> 
   <td>&gt;=455.23</td> 
   <td>&gt;=456.38</td> 
  </tr> 
  <tr> 
   <td>CUDA 11.0.3 Update 1</td> 
   <td>&gt;= 450.51.06</td> 
   <td>&gt;= 451.82</td> 
  </tr> 
  <tr> 
   <td>CUDA 11.0.2 GA</td> 
   <td>&gt;= 450.51.05</td> 
   <td>&gt;= 451.48</td> 
  </tr> 
  <tr> 
   <td>CUDA 11.0.1 RC</td> 
   <td>&gt;= 450.36.06</td> 
   <td>&gt;= 451.22</td> 
  </tr> 
  <tr> 
   <td>CUDA 10.2.89</td> 
   <td>&gt;= 440.33</td> 
   <td>&gt;= 441.22</td> 
  </tr> 
  <tr> 
   <td>CUDA 10.1 (10.1.105 general release, and updates)</td> 
   <td>&gt;= 418.39</td> 
   <td>&gt;= 418.96</td> 
  </tr> 
  <tr> 
   <td>CUDA 10.0.130</td> 
   <td>&gt;= 410.48</td> 
   <td>&gt;= 411.31</td> 
  </tr> 
  <tr> 
   <td>CUDA 9.2 (9.2.148 Update 1)</td> 
   <td>&gt;= 396.37</td> 
   <td>&gt;= 398.26</td> 
  </tr> 
  <tr> 
   <td>CUDA 9.2 (9.2.88)</td> 
   <td>&gt;= 396.26</td> 
   <td>&gt;= 397.44</td> 
  </tr> 
  <tr> 
   <td>CUDA 9.1 (9.1.85)</td> 
   <td>&gt;= 390.46</td> 
   <td>&gt;= 391.29</td> 
  </tr> 
  <tr> 
   <td>CUDA 9.0 (9.0.76)</td> 
   <td>&gt;= 384.81</td> 
   <td>&gt;= 385.54</td> 
  </tr> 
  <tr> 
   <td>CUDA 8.0 (8.0.61 GA2)</td> 
   <td>&gt;= 375.26</td> 
   <td>&gt;= 376.51</td> 
  </tr> 
  <tr> 
   <td>CUDA 8.0 (8.0.44)</td> 
   <td>&gt;= 367.48</td> 
   <td>&gt;= 369.30</td> 
  </tr> 
  <tr> 
   <td>CUDA 7.5 (7.5.16)</td> 
   <td>&gt;= 352.31</td> 
   <td>&gt;= 353.66</td> 
  </tr> 
  <tr> 
   <td>CUDA 7.0 (7.0.28)</td> 
   <td>&gt;= 346.46</td> 
   <td>&gt;= 347.62</td> 
  </tr> 
 </tbody> 
</table>

> 此时我们已经确定了当前环境为 64位 Ubuntu 16.04，显卡 1080Ti
> 
> 目标 Cuda 10.2 以上,因此需要驱动版本 >= 440.33

### 下载驱动 ###

*  官网下载 https://www.nvidia.cn/geforce/drivers/

![532e0b6e135ef6e14f1fbfd5c807b4bf.png][]

*  在搜索列表中选择合适自己目标的驱动版本(也不建议选择最新的,应用开发有可能跟不上,够用就好)
 *  我下载了 `NVIDIA-Linux-x86_64-450.80.02.run`

### 安装驱动 ###

> 需要强调的是我现在的安装环境是Linux服务器,没有图形界面,使用图形界面的同志不要按照此方案安装.

*  卸载已有的驱动

sudo /usr/bin/nvidia-uninstall
    sudo apt-get --purge remove nvidia-*

*  安装新驱动

sudo chmod a+x NVIDIA-Linux-x86_64-450.80.02.run
    sudo ./NVIDIA-Linux-x86_64-450.80.02.run -no-x-check -no-nouveau-check -no-opengl-files

> –no-opengl-files 只安装驱动文件,不安装OpenGL文件
> 
> –no-x-check 安装驱动时不检查X服务
> 
> –no-nouveau-check 安装驱动时不检查nouveau

### 查看驱动更新结果 ###

$ nvidia-smi

![6fc2f8c3d9d30aad680ff9ccfbf34e96.png][]

*  更新成功。

### 参考资料 ###

*  https://blog.csdn.net/qq\_30163461/article/details/80314630?utm\_medium=distribute.pc\_relevant.none-task-blog-BlogCommendFromBaidu-2.control&depth\_1-utm\_source=distribute.pc\_relevant.none-task-blog-BlogCommendFromBaidu-2.control
 *  https://www.cnblogs.com/haiyang21/p/12699593.html

[532e0b6e135ef6e14f1fbfd5c807b4bf.png]: /images/20221120/669ae8b648a644fe90e8f15505bd4846.png
[6fc2f8c3d9d30aad680ff9ccfbf34e96.png]: /images/20221120/d13601ff39314bb5a35871b67b1184c3.png