devbox
知新_RL
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

Ubuntu Server 20.04 系统安装（四）：深度学习 GPU 环境配置（CUDA12.2驱动+CUDA Toolkit 12.2+cuDNN v8.9.7）_ubuntu cuda12

作者：知新_RL | 2024-02-22 05:04:43

赞

踩

ubuntu cuda12

引言

本文以 Ubuntu 20.04 操作系统为例，演示如何配置深度学习 GPU 环境。

一、NVIDIA 显卡驱动的安装

访问如下网址

https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=20.04&target_type=runfile_local
在这里插入图片描述

下载推荐的cuda并安装

复制箭头所指处的命令到命令行

wget https://developer.download.nvidia.com/compute/cuda/12.2.0/local_installers/cuda_12.2.0_535.54.03_linux.run

必须先关图形界面（重要）

如果不关，运行下面安装的命令之后会报错，装不上

systemctl isolate multi-user.target

关闭图形界面之后会重启，输入如下命令

安装驱动

sudo sh cuda_12.2.0_535.54.03_linux.run

稍等一会
在弹出的安装界面中选“continue”，如果选了会跳出安装，就说明安装失败，给了失败日志的路径，自己查看原因，一般是gcc版本问题，

gcc降级

  1. sudo apt-get install gcc-7 g++-7
  2.  
  3. sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 9
  4. sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 1
  5.  
  6. sudo update-alternatives --display gcc     //查看默认版本
  7.  
  8. sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-7 9
  9. sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-9 1
  10.  
  11. sudo update-alternatives --display g++

如果没问题继续安装：
在这里插入图片描述
cuda安装包是自带显卡驱动的，所以这一步按空格去掉安装显卡驱动的选项，然后选择install
在这里插入图片描述

开启图形界面

sudo systemctl start graphical.target

二、安装 CUDA Toolkit 12.2

准备

开始安装之前，需要修改 apt-get 的镜像源，不然国内下载速度很慢。我们这里以阿里源为例演示如何修改。

  1. 备份官方源
sudo mv /etc/apt/sources.list /etc/apt/sources.list.bak

安装vim：

sudo apt-get install vim

运行 sudo vim /etc/apt/sources.list 打开文件，并按 i 插入以下内容，按 ESC 后再按 :wq 保存退出

sudo vim /etc/apt/sources.list
  1. deb http://mirrors.aliyun.com/ubuntu/ focal main restricted universe multiverse
  2. deb-src http://mirrors.aliyun.com/ubuntu/ focal main restricted universe multiverse
  3.  
  4. deb http://mirrors.aliyun.com/ubuntu/ focal-security main restricted universe multiverse
  5. deb-src http://mirrors.aliyun.com/ubuntu/ focal-security main restricted universe multiverse
  6.  
  7. deb http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted universe multiverse
  8. deb-src http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted universe multiverse
  9.  
  10. # deb http://mirrors.aliyun.com/ubuntu/ focal-proposed main restricted universe multiverse
  11. # deb-src http://mirrors.aliyun.com/ubuntu/ focal-proposed main restricted universe multiverse
  12.  
  13. deb http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
  14. deb-src http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
  1. 更新数据源列表
sudo apt-get update

安装参考 NVIDIA 官网提供的教程。

Quick Start Guide 12.3 documentation

  1. wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
  2. sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
  3. wget https://developer.download.nvidia.com/compute/cuda/12.3.1/local_installers/cuda-repo-ubuntu2004-12-3-local_12.3.1-545.23.08-1_amd64.deb
  4. sudo dpkg -i cuda-repo-ubuntu2004-12-3-local_12.3.1-545.23.08-1_amd64.deb
  5. sudo cp /var/cuda-repo-ubuntu2004-12-3-local/cuda-*-keyring.gpg /usr/share/keyrings/
  6. sudo apt-get update
  7. sudo apt-get -y install cuda-toolkit-12-3

sudo reboot # 非必需

执行完上述安装操作后，需要配置环境变量，让该系统下的所有用户都能使用：

  1. 打开全局配置文件
sudo vi /etc/profile
  1. 在文件末尾添加如下内容后保存退出。
  1. export PATH=/usr/local/cuda-12.3/bin${PATH:+:${PATH}}
  2. export LD_LIBRARY_PATH=/usr/local/cuda-12.3/lib64\${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

  1. 激活环境变量
source /etc/profile
  1. 运行 nvcc -V 来校验安装是否成功，若正确显示版本号则说明安装成功
  1. $ nvcc -V
  2. nvcc: NVIDIA (R) Cuda compiler driver
  3. Copyright (c) 2005-2023 NVIDIA Corporation
  4. Built on Fri_Nov__3_17:16:49_PDT_2023
  5. Cuda compilation tools, release 12.3, V12.3.103
  6. Build cuda_12.3.r12.3/compiler.33492891_0
  7.

二、安装 cuDNN v8.9.7

安装参考 NVIDIA 官方提供的教程。

  1. 从官网下载合适版本的离线安装包


2. 开始安装

  1. sudo dpkg -i cudnn-local-repo-ubuntu2004-8.9.7.29_1.0-1_amd64.deb
  2. sudo cp /var/cudnn-local-repo-ubuntu2004-8.9.7.29/cudnn-local-30472A84-keyring.gpg /usr/share/keyrings/
  3. cd /var/cudnn-local-repo-ubuntu2004-8.9.7.29/
  4. sudo dpkg -i libcudnn8_8.9.7.29-1+cuda12.2_amd64.deb
  5. sudo dpkg -i libcudnn8-dev_8.9.7.29-1+cuda12.2_amd64.deb
  6. sudo dpkg -i libcudnn8-samples_8.9.7.29-1+cuda12.2_amd64.deb
  7.
  1. 校验安装是否成功（最终输出 Test passed! 则说明安装成功）
  1. cp -r /usr/src/cudnn_samples_v8/ $HOME
  2. cd  $HOME/cudnn_samples_v8/mnistCUDNN
  3. make clean && make
  4. ./mnistCUDNN
  5.

校验安装时如果遇到下述错误：

  1. ```bash
  2. test.c:1:10: fatal error: FreeImage.h: No such file or directory
  3.     1 | #include "FreeImage.h"
  4.       |          ^~~~~~~~~~~~~
  5. compilation terminated.
  6. >>> WARNING - FreeImage is not set up correctly. Please ensure FreeImage is set up correctly. <<<

可通过如下方式解决，然后重新进行校验

sudo apt-get -y install  libfreeimage3 libfreeimage-dev

重启后验证通过

sudo reboot

验证通过：

  1. wyr@ubuntu2004w1:~/cudnn_samples_v8/mnistCUDNN$ ./mnistCUDNN
  2. Executing: mnistCUDNN
  3. cudnnGetVersion() : 8401 , CUDNN_VERSION from cudnn.h : 8401 (8.4.1)
  4. Host compiler version : GCC 9.4.0
  5.  
  6. There are 2 CUDA capable devices on your machine :
  7. device 0 : sms 56  Capabilities 6.0, SmClock 1328.5 Mhz, MemSize (Mb) 16280, MemClock 715.0 Mhz, Ecc=1, boardGroupID=0
  8. device 1 : sms 56  Capabilities 6.0, SmClock 1328.5 Mhz, MemSize (Mb) 16280, MemClock 715.0 Mhz, Ecc=1, boardGroupID=1
  9. Using device 0
  10.  
  11. Testing single precision
  12. Loading binary file data/conv1.bin
  13. Loading binary file data/conv1.bias.bin
  14. Loading binary file data/conv2.bin
  15. Loading binary file data/conv2.bias.bin
  16. Loading binary file data/ip1.bin
  17. Loading binary file data/ip1.bias.bin
  18. Loading binary file data/ip2.bin
  19. Loading binary file data/ip2.bias.bin
  20. Loading image data/one_28x28.pgm
  21. Performing forward propagation ...
  22. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  23. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory
  24. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  25. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 57600 memory
  26. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 178432 memory
  27. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 184784 memory
  28. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 2057744 memory
  29. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  30. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  31. Testing cudnnFindConvolutionForwardAlgorithm ...
  32. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.024864 time requiring 0 memory
  33. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.025152 time requiring 0 memory
  34. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.066912 time requiring 57600 memory
  35. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.097024 time requiring 178432 memory
  36. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.144352 time requiring 2057744 memory
  37. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.275520 time requiring 184784 memory
  38. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  39. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  40. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  41. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory
  42. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  43. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 128000 memory
  44. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 4656640 memory
  45. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 1433120 memory
  46. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 2450080 memory
  47. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  48. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  49. Testing cudnnFindConvolutionForwardAlgorithm ...
  50. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.062560 time requiring 0 memory
  51. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.062880 time requiring 0 memory
  52. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.107712 time requiring 2450080 memory
  53. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.114528 time requiring 1433120 memory
  54. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.116704 time requiring 128000 memory
  55. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.130944 time requiring 4656640 memory
  56. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  57. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  58. Resulting weights from Softmax:
  59. 0.0000000 0.9999399 0.0000000 0.0000000 0.0000561 0.0000000 0.0000012 0.0000017 0.0000010 0.0000000 
  60. Loading image data/three_28x28.pgm
  61. Performing forward propagation ...
  62. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  63. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory
  64. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  65. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 57600 memory
  66. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 178432 memory
  67. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 184784 memory
  68. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 2057744 memory
  69. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  70. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  71. Testing cudnnFindConvolutionForwardAlgorithm ...
  72. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.024512 time requiring 0 memory
  73. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.040768 time requiring 0 memory
  74. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.052480 time requiring 57600 memory
  75. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.067328 time requiring 2057744 memory
  76. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.069856 time requiring 178432 memory
  77. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.077664 time requiring 184784 memory
  78. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  79. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  80. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  81. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory
  82. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  83. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 128000 memory
  84. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 4656640 memory
  85. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 1433120 memory
  86. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 2450080 memory
  87. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  88. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  89. Testing cudnnFindConvolutionForwardAlgorithm ...
  90. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.062336 time requiring 0 memory
  91. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.064128 time requiring 0 memory
  92. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.077600 time requiring 1433120 memory
  93. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.079904 time requiring 2450080 memory
  94. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.113632 time requiring 128000 memory
  95. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.123968 time requiring 4656640 memory
  96. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  97. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  98. Resulting weights from Softmax:
  99. 0.0000000 0.0000000 0.0000000 0.9999288 0.0000000 0.0000711 0.0000000 0.0000000 0.0000000 0.0000000 
  100. Loading image data/five_28x28.pgm
  101. Performing forward propagation ...
  102. Resulting weights from Softmax:
  103. 0.0000000 0.0000008 0.0000000 0.0000002 0.0000000 0.9999820 0.0000154 0.0000000 0.0000012 0.0000006 
  104.  
  105. Result of classification: 1 3 5
  106.  
  107. Test passed!
  108.  
  109. Testing half precision (math in single precision)
  110. Loading binary file data/conv1.bin
  111. Loading binary file data/conv1.bias.bin
  112. Loading binary file data/conv2.bin
  113. Loading binary file data/conv2.bias.bin
  114. Loading binary file data/ip1.bin
  115. Loading binary file data/ip1.bias.bin
  116. Loading binary file data/ip2.bin
  117. Loading binary file data/ip2.bias.bin
  118. Loading image data/one_28x28.pgm
  119. Performing forward propagation ...
  120. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  121. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 5632 memory
  122. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  123. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 28800 memory
  124. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 178432 memory
  125. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 184784 memory
  126. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 2057744 memory
  127. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  128. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  129. Testing cudnnFindConvolutionForwardAlgorithm ...
  130. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.026944 time requiring 0 memory
  131. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.052864 time requiring 28800 memory
  132. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.064928 time requiring 178432 memory
  133. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.073376 time requiring 2057744 memory
  134. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.093312 time requiring 184784 memory
  135. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.109312 time requiring 5632 memory
  136. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  137. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  138. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  139. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 1433120 memory
  140. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 4656640 memory
  141. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 2450080 memory
  142. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 2000 memory
  143. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  144. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 0 memory
  145. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  146. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  147. Testing cudnnFindConvolutionForwardAlgorithm ...
  148. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.087264 time requiring 2450080 memory
  149. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.088320 time requiring 1433120 memory
  150. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.088864 time requiring 0 memory
  151. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.097152 time requiring 2000 memory
  152. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.122624 time requiring 4656640 memory
  153. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.195200 time requiring 0 memory
  154. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  155. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  156. Resulting weights from Softmax:
  157. 0.0000001 1.0000000 0.0000001 0.0000000 0.0000563 0.0000001 0.0000012 0.0000017 0.0000010 0.0000001 
  158. Loading image data/three_28x28.pgm
  159. Performing forward propagation ...
  160. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  161. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 5632 memory
  162. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  163. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 28800 memory
  164. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 178432 memory
  165. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 184784 memory
  166. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 2057744 memory
  167. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  168. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  169. Testing cudnnFindConvolutionForwardAlgorithm ...
  170. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.026272 time requiring 0 memory
  171. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.047712 time requiring 28800 memory
  172. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.068864 time requiring 2057744 memory
  173. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.069856 time requiring 5632 memory
  174. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.075904 time requiring 184784 memory
  175. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.097664 time requiring 178432 memory
  176. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  177. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  178. Testing cudnnGetConvolutionForwardAlgorithm_v7 ...
  179. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 1433120 memory
  180. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 4656640 memory
  181. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 2450080 memory
  182. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 2000 memory
  183. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory
  184. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 0 memory
  185. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  186. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  187. Testing cudnnFindConvolutionForwardAlgorithm ...
  188. ^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 0.079712 time requiring 2450080 memory
  189. ^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.079808 time requiring 2000 memory
  190. ^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.083232 time requiring 1433120 memory
  191. ^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.088128 time requiring 0 memory
  192. ^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.117088 time requiring 4656640 memory
  193. ^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.185120 time requiring 0 memory
  194. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory
  195. ^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory
  196. Resulting weights from Softmax:
  197. 0.0000000 0.0000000 0.0000000 1.0000000 0.0000000 0.0000714 0.0000000 0.0000000 0.0000000 0.0000000 
  198. Loading image data/five_28x28.pgm
  199. Performing forward propagation ...
  200. Resulting weights from Softmax:
  201. 0.0000000 0.0000008 0.0000000 0.0000002 0.0000000 1.0000000 0.0000154 0.0000000 0.0000012 0.0000006 
  202.  
  203. Result of classification: 1 3 5
  204.  
  205. Test passed!
  206. wyr@ubuntu2004w1:~/cudnn_samples_v8/mnistCUDNN$

安装CUDA后变成了桌面版的了，让我很纳闷，只好一个一个的卸载

  1. # 卸载gnome
  2. sudo apt-get remove gnome -y
  3. # 卸载 gnome-shell
  4. sudo apt-get remove gnome-shell -y
  5. # 彻底卸载删除gnome的相关配置文件
  6. sudo apt-get purge gnome -y
  7. # 卸载臃肿的服务组件
  8. sudo apt-get remove snapd -y
  9. # 卸载依赖组件
  10. sudo apt-get autoremove -y
  11. # 清理安装gnome时候留下的缓存程序软件包
  12. sudo apt-get autoclean
  13. sudo apt-get clean
  14. # 重启系统
  15. sudo reboot 
  16. sudo shutdown -r now

GPUstat

这是一个可以替换 nvidia-smi 查看显卡信息的一个工具。可以非常清楚地看到各个显卡的温度、利用率、正在使用的用户以及各用户占用的显存等。

在这里插入图片描述

使用如下命令安装：

  1. sudo apt install python3-pip
  2. sudo pip install gpustat

用法非常简单，记住下面这一条命令就行：

gpustat -i

NetSM

这是一个跨平台的命令行网速监测工具，可以看到服务器实时网速显示。

在这里插入图片描述

通过如下命令安装：

  1. sudo apt install python3-pip
  2. sudo pip install netsm

用法非常简单，记住下面这一条命令就行：

netsm show

Tmux

跑耗时较长的任务时，可以打开 tmux 将任务挂起，即使你关闭了窗口，任务仍然在后台运行。

sudo apt-get install tmux

Axel

区别于 wget 单线程下载的多线程下载工具。用法很简单，通过 -n 参数指定线程数即可，例如：axel -n 10 下载链接。

sudo apt-get install axel

提示

上述配置了三个深度学习 GPU 环境，分别是 PyTorch，TensorFlow 2 和 TensorFlow 1。且都是通过管理员来配置的。其他普通用户使用时，仅需使用 conda 在新建环境时克隆即可，下面给一个例子。

例如，普通用户要使用 PyTorch 环境，但他同时还需要安装 requests 包，直接在名为 pytorch 的环境里面是没有办法安装，因为缺少写入权限。因此可以通过如下克隆的方式安装：

  1. conda create -n pytorch2 --clone pytorch
  2. conda activate pytorch2
  3. pip install requests

这样既节省了服务器的存储空间，又避免了重复安装环境的烦恼。

当然你不克隆，直接激活使用也是可以的：

  1. conda activate pytorch
  2. pip install requests

因为默认会安装 requests 到用户家目录下，不会和其他用户的环境产生冲突。但如果环境多了，就不太容易区分了。

参考

简易记录：安装GPU驱动，CUDA和cuDNN
使用ubuntu 20.04 进行机器学习配置环境 必须做的事情
【Linux】在Ubuntu上安装和使用miniconda
Administering a multi-user conda installation
Start Locally | PyTorch
Anaconda 镜像使用帮助
使用 pip 安装 TensorFlow

ubuntu20.04安装12.2版本的cuda_cuda12.2-CSDN博客

【Linux】Ubuntu 20.04 深度学习 GPU 环境配置（CUDA Toolkit 11.7 + cuDNN v8.4.1）_Xavier Jiezou的博客-CSDN博客

ubuntu20.04 TLS Server升级后安装了gnome桌面环境_u013250861的博客-CSDN博客

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/知新_RL/article/detail/128612
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号