centos7安装tensorflow_gpu完整教程

Love The Way You Lie 2023-05-30 03:50 21阅读 0赞

**配置信息**

丽台K2200+显卡驱动440.31+cuda9.0+cudnn7.6.2+python3.6+tensorflow\_gpu1.12.0（试试这个）

**1****、安装显卡驱动**

1.1查看显卡型号，命令：lspci  | grep -i vga，

1.2根据显卡型号下载显卡驱动，下载地址：[https://www.nvidia.cn/Download/index.aspx?lang=cn][https_www.nvidia.cn_Download_index.aspx_lang_cn]

1.3安装显卡驱动

（1）安装依赖环境：yum install gcc kernel-delve -y（要确保内核版本号和内核源码版本号一致，查看内核版本：ls /boot | grep vmlinu，查看源码包版本：rpm -aq | grep kernel-devel）

（2）屏蔽系统默认的mouveau，root用户打开gedit /lib/modprobe.d/dist-blacklist.conf，注释掉\#blacklist nvidiafb，添加blacklist nouveau和options nouveau modeset=0（查看mouveau是否禁用：lsmod | grep nouveau）

（3）重建initramfs image，两条命令如下：

mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r).img.bak

dracut /boot/initramfs-$(uname -r).img $(uname -r)

（4）修改运行级别为文本模式并重启：systemctl set-default multi-user.target和reboot

（5）进入下载的驱动所在目录，修改执行权限：chmod +x NVIDIA-Linux-x86\_64-346.47.run，执行安装：./NVIDIA-Linux-x86\_64-346.47.run

（6）回到图像模式并重启systemctl set-default graphical.target和reboot

（7）验证nvidia-smi

**2****、安装cuda**

2.1下载cuda，下载地址：[https://developer.nvidia.com/cuda-downloads(cuda10)，https://developer.nvidia.com/cuda-90-download-archive(cuda9)][https_developer.nvidia.com_cuda-downloads_cuda10_https_developer.nvidia.com_cuda-90-download-archive_cuda9]，可以用wget也可以先下载下来)

2.2修改运行级别为文本模式并重启：systemctl set-default multi-user.target和reboot

2.3 进入下载的驱动所在目录，修改执行权限chmod +x cuda\_8.0.44\_linux.run，执行：./cuda\_8.0.44\_linux.run

2.4安装中的设置

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70][]

2.5回到图像模式并重启systemctl set-default graphical.target和reboot

2.6设置环境变量：在/etc/profile文件中添加三行：

export PATH=/usr/local/cuda/bin$\{PATH:+:$\{PATH\}\}

export LD\_LIBRARY\_PATH=/usr/local/cuda/lib64$\{LD\_LIBRARY\_PATH:+:$\{LD\_LIBRARY\_PATH\}\}

export CUDA\_HOME=/usr/local/cuda

，并执行source /etc/profile使其立即生效。（注意这里的路径就是cuda，而不是cuda-9.0）

2.6验证 nvcc -V

**3****、安装cudnn**

3.1下载cudnn，下载地址：[https://developer.nvidia.com/rdp/cudnn-archive][https_developer.nvidia.com_rdp_cudnn-archive]

3.2解压tar -xvf cudnn-8.0-linux-x64-v6.0.tgz -C /usr/local/  （貌似只解压，不知道徐不需要复制，复制参考[https://www.jianshu.com/p/a201b91b3d96][https_www.jianshu.com_p_a201b91b3d96]）

**4****、安装python3.6**

4.1解决基础依赖yum install -y openssl-devel openssl-static zlib-devel lzma tk-devel xz-devel bzip2-devel ncurses-devel gdbm-devel readline-devel sqlite-devel gcc libffi-devel

4.2下载python：wget [https://www.python.org/ftp/python/3.6.0/Python-3.6.0.tgz][https_www.python.org_ftp_python_3.6.0_Python-3.6.0.tgz]

4.2解压并迁移tar -xvf Python-3.6.0.tgz 和mv Python-3.7.0 /usr/local

4.3执行配置文件并编译cd Python-3.6.0和 ./configure和make&&make install

4.4替换之前版本的python

建立软连接mv /usr/bin/python /usr/bin/python\_bak和ln -s /usr/local/Python-3.6.0/python /usr/bin/python，输入python看是不是python3.6.0（替换之前版本后，因为系统默认是python2.x，所以需要把一些文件中python 改为python2.7）

4.5、安装pip并修改源（可以不做，直接在pip3命令上加参数即可）：

yum install python-pip和pip install --upgrade pip（试一下pip和pip3两个命令，貌似也都需要修改源）

\[global\]

index-url = http://mirrors.aliyun.com/pypi/simple/

\[install\]

trusted-host=mirrors.aliyun.com

![20191107103150781.png][]

**5****、pip安装tesnsorflow**，

pip3 install tensorflow-gpu==1.12.0

不配置pip源可以使用pip3 install tensorflow-gpu==1.12.0 -i [http://pypi.douban.com/simple --trusted-host pypi.douban.com][http_pypi.douban.com_simple --trusted-host pypi.douban.com]

**参考：**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70 1][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70 2][]

[https_www.nvidia.cn_Download_index.aspx_lang_cn]: https://www.nvidia.cn/Download/index.aspx?lang=cn
[https_developer.nvidia.com_cuda-downloads_cuda10_https_developer.nvidia.com_cuda-90-download-archive_cuda9]: https://developer.nvidia.com/cuda-downloads%28cuda10%29%EF%BC%8Chttps:/developer.nvidia.com/cuda-90-download-archive%28cuda9%29
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20191107103130350.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz,size_16,color_FFFFFF,t_70
[https_developer.nvidia.com_rdp_cudnn-archive]: https://developer.nvidia.com/rdp/cudnn-archive
[https_www.jianshu.com_p_a201b91b3d96]: https://www.jianshu.com/p/a201b91b3d96
[https_www.python.org_ftp_python_3.6.0_Python-3.6.0.tgz]: https://www.python.org/ftp/python/3.6.0/Python-3.6.0.tgz
[20191107103150781.png]: https://img-blog.csdnimg.cn/20191107103150781.png
[http_pypi.douban.com_simple --trusted-host pypi.douban.com]: http://pypi.douban.com/simple%20--trusted-host%20pypi.douban.com
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20191107103201896.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/2019110710321058.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM0NTYyMDkz,size_16,color_FFFFFF,t_70