1 Ubuntu 16.04 LTS安装

由于是作为tensorflow的训练机，所以稳定性考虑更重要，选择最新的LTS版本。由于Intel第七代CPU过于新，而Ubuntu对其支持不好，导致安装光盘根本无法启动安装，解决方案：

• 用安装U盘或者其他进入到GRUB启动选项
• 用上下键选择安装选项，然后按键盘的e，进入启动脚本编辑环境
• 找到quiet splash，替换为nomodeset
• 然后正常安装系统
• 完成后给系统设置固定IP地址

2 Samba服务器

安装samba的目的是为了便于文件交互，方面后面的操作和以后远程代码copy

2.1 Install Samba

apt get update
apt get install samba

2.2 Modify Config

vi /etc/samba/smb.confsecurity = user                 # add
[homes]                         # modifycomment = Home Directories   # modifybrowseable = no              # modifypath = /home/user/work       # add, edit user to your user nameread only = no              # modify

2.3 Add Samba passwd

pdbedit -a -u username

2.4 service restart

/etc/init.d/samba restart

3 ssh证书登录

3.1 Windows

3.1.1 key generate

• use puttygen generate rsa key, and public key
• save private key
• copy Pub key for openSSH, and save as authorized_keys
• use samba to send authorized_keys to server

3.1.2 server setup

cd ~
mkdir .ssh
cd .ssh
cat xxx/authorized_keys >> authorized_keys

3.1.3 putty client setup

• Session -> IP & Port: 192.168.1.27 & 22
• Connection -> Data -> user name: file
• Connection -> SSH -> Auth: load the private key
• Session -> Saved Sessions: save all those setting as XXX

3.2 Linux

3.2.1 key generate

ssh-keygen -t rsa -b 1024
ssh-keygen -y -f ~/.ssh/id_rsa >> keytext
sudo ssh-keygen -f "/home/user/.ssh/known_hosts" -R 192.168.1.89 # add server to trust list

3.2.2 server setup

cat keytext >> authorized_keys

3.2.3 client login

ssh user@192.168.1.88

3.3 More safe login setup of server

close password login port using ssh:

vim /etc/ssh/sshd_configPasswordAuthentication no                          ### change

4 安装python3下GPU版本的tensorflow

apt-get update
apt-get upgradeapt-get install python3-pip
pip3 install --upgrade pippip3 install pylint numpy scipy matplotlib
pip3 install tensorflow-gpu

5 安装CUDA（以root用户操作）

5.1 CUDA下载

• CUDA官方下载地址下载CUDA安装包
• CUDNN官方下载地址下载CNN安装包

5.2安装

5.2.1 创建系统默认显卡驱动nouveau的黑名单

vim  /etc/modprobe.d/blacklist-nouveau.confblacklist nouveau
options nouveau modeset=0

5.2.2 重新生成initramfs内核

sudo update-initramfs -u

5.2.3 系统重启

reboot

5.2.4 ssh登录tensorflow PC，kill x-server进程

service lightdm stop

5.2.5 CUDA安装

进入CUDA下载目录，然后执行如下命令进行安装（我下载的文件名为cuda_8.0.61_375.26_linux.run）

chmod +x cuda_8.0.61_375.26_linux.run
./cuda_8.0.61_375.26_linux.run

安装的时候全部默认安装，除了OpenGL

5.2.6 CUDNN安装

进入CUDNN下载目录，然后执行如下命令安装（我的下载文件名为 cudnn-8.0-linux-x64-v5.1.solitairetheme8 ）

tar -zxvf cudnn-8.0-linux-x64-v5.1.solitairetheme8cd cuda
mv ./lib64/* /usr/local/cuda/lib64/ 
mv include/cudnn.h /usr/local/cuda/include/cd  /usr/local/cuda/lib64/
chmod +r  libcudnn.so.5.1.10
ln -sf libcudnn.so.5.1.10 libcudnn.so.5
ln -sf libcudnn.so.5 libcudnn.so

5.2.7 环境配置

• 在/etc/profile中添加CUDA环境变量

vim /etc/profilePATH=/usr/local/cuda/bin:$PATH        # add at the end of profile
export PATH                           # add at the end of profile

• 刷新环境变量

source  /etc/profile

5.2.8 添加lib库路径

• 在/etc/ld.so.conf.d/中新建一个cuda.conf文件

vim /etc/ld.so.conf.d/cuda.conf/usr/local/cuda/lib64 
/lib

• 更新系统lib文件的链接路径

ldconfig  -v

5.2.9 确认安装

#!/usr/bin/python3import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
print(sess.run(hello))
a = tf.constant(10)
b = tf.constant(32)
print(sess.run(a+b))

如果输出结果中有如下信息，则表示CUDA配置成功

I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.8.0 locally