Configure CentOS7 GPU environment

basic configuration##

1. Install wget first
   yum -y install wget

2. Create your own folder
   mkdir sunyan

Install Anoconda

1. Download Anoconda. Note that the latest Anoconda is version 3.7, but tensorflow has not been updated to 3.7. Here we install the python 3.6 version released in May 2018.
   wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-5.2.0-Linux-x86_64.sh

2. Install bzip2
   yum install -y bzip2

3. Install Anoconda, yes all the way.
   bash Anaconda3-5.2.0-Linux-x86_64.sh

4. Enable configuration
   source .bashrc

5. Enter Python to see if the configuration is correct

Conda installs TesnorFlow GPU version##

1. Configure domestic conda source. Because cudnn and cuda are very large and conda is abroad, it is easy to cause download interruption. Here we configure the mirror image of the Chinese University of Science and Technology in China.
   conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/main/

2. installation
   conda install tensorflow-gpu

My cuda version is 9.0, cudnn version is 7.1.2, and tensorflow-gpu version is 1.9.0.

Install NVIDIA driver##

1. Install gcc
   yum -y install gcc-c++

2. Install Open JDK
   yum install java-1.8.0-openjdk java-1.8.0-openjdk-devel

3. Check the graphics card driver and model
   sudo rpm --import https://www.elrepo.org/RPM-GPG-KEY-elrepo.org sudo rpm -Uvh http://www.elrepo.org/elrepo-release-7.0-2.el7.elrepo.noarch.rpm sudo yum install nvidia-detect

Shown here is Tesla P4, 410.66 NVIDIA driver

1. Go to Official Website to download the corresponding driver
   wget http://cn.download.nvidia.com/tesla/384.145/NVIDIA-Linux-x86_64-384.145.run

2. Shield the default nouveau, use the su command to switch to the root user, and then modify the /etc/modprobe.d/blacklist.conf file, if the system does not have this file, you need to create a new one.
   su root echo -e "blacklist nouveau\noptions nouveau modeset=0" > /etc/modprobe.d/blacklist.conf

Then check the contents of the file:

cat /etc/modprobe.d/blacklist.conf

1. Rebuild the initramfs image
   mv /boot/initramfs-(uname -r).img /boot/initramfs-(uname -r).img.bak dracut /boot/initramfs-$(uname -r).img $(uname -r)

2. Modify the run level to text mode

systemctl set-default multi-user.target

1. Restart, log in as root user, you need to wait here, about 1 minute.
   sudo iptables stop reboot

2. Check whether nouveau has been disabled. If no relevant content is displayed, it means it has been disabled.
   ls mod | grep nouveau

3. Install kenel-devel
   wget ftp://ftp.riken.jp/Linux/cern/centos/7/updates/x86_64/Packages/kernel-devel-3.10.0-693.17.1.el7.x86_64.rpm yum install kernel-devel-3.10.0-693.17.1.el7.x86_64.rpm

4. Install NVIDIA driver, just enter all the way
   chmod +x NVIDIA-Linux-x86_64-384.145.run sh NVIDIA-Linux-x86_64-384.145.run

test##

1. nvidia-smi test and TensorFlow test, just display GPU information. It doesn't matter if the warning is reported, it is a warning from TF itself, and it can also be solved by pip install h5py==2.8.0rc1. nvidia-smi python3 import tensorflow as tf sess = tf.Session()

1. Replace PyPi as a domestic source
   pip install pqi pqi use tuna pqi show

2. Replace conda source with domestic source
   conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/main/ conda config --add channels https://mirrors.ustc.edu.cn/anaconda/pkgs/free/ conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/conda-forge/ conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/msys2/ conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/bioconda/ conda config --add channels https://mirrors.ustc.edu.cn/anaconda/cloud/menpo/ conda config --set show_channel_urls yes