How to install tensorflow with CUDA and cuDNN on Ubuntu 16.04

Environment

python3.5+tensorflow-gpu1.5 + cuda9.0 + cudnn9.0

installation steps

Disable open source nouveau driver

sudo vim /etc/modprobe.d/disable-nouveau.conf

Add the following two lines at the end.

blacklist nouveauoptions 
nouveau modeset=0

Then execute the following command to make the configuration take effect.

sudo ldconfig
sudo update-initramfs -u

Verify successful disabling of nouveau.

sudo lspci | grep nouveauinstall

Disable success if there is no output.

Install the Linux4.4.0 kernel

sudo apt-get install linux-headers-4.4.0-93-generic
sudo apt-get install inux-image-4.4.0-93-generic

Restart the computer, enter the Ubuntu advanced options in the boot menu, and select the linux 4.4.0 kernel to start the system.
Press ctrl+alt+f1 on the login interface to enter tty1 mode and close the desktop.

sudo service lightdm stop

Install CUDA9.0

sudo ./cuda_9.0.*_linux.run

Install cuDNN9.0

Switch to the directory where the cuDNN Tar file is located and unzip it.

tar -xzvf cudnn-9.0-linux-x64-v7.tgz

Copy the following files to the CUDA directory.

sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h

Environment variable configuration

sudo vi ~/.bashrc (console valid) or /etc/profile

Write the following to the end of the file.

export PATH=PATH:/usr/local/cuda-9.0/bin
export LD_LIBRARY_PATH=LD_LIBRARY_PATH:/usr/local/cuda-9.0/lib64

After saving, update the file.

source ~/.bashrc

Install NVIDIA Driver

Restart, also press ctrl + alt + f1 on the login interface to enter tty1, and install the NVIDIA graphics card driver.

sudo sh ./NVIDIA-Linux-x86_64-38* -no-x-check -no-nouveau-check -no-opengl-files

Test

Restart, enter the graphical interface. Test if the installation is successful.

python3
import numpy import tensorflow as tf 
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a') 
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b') 
c = tf.matmul(a, b) 
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)) 
print(sess.run(c)) 
exit()