It's all tear and blood. Notes after reinstalling the system (Note that I didn't check it, the graphics card settings can refer to other posts)

Install CUDA

Download address, official website or Baidu cloud, Baidu cloud address

https://pan.baidu.com/s/1ZjI3LDlLpRf_NSVsrj7WSw

iqqx

Official website download failed, not to mention for the moment.

Document:

Enter the directory where the deb file is located, command line operation

sudo dpkg -i cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64.deb
sudo apt-key add /var/cuda-repo-9-0-local/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda

Click yes.. runfile installed before is probably whether to install the nvidia driver, anyway, I did not install the nvidia driver, so all the way yes. If the nvidia driver is installed beforehand, select no as the driver option.
Add some libraries

sudo apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev

restart

nvidia-smi

command line

ls /dev/nvidia*

If there is uvm, the installation is successful.

Setting environment variables

sudo gedit /etc/profile

If something goes wrong, see the next post.

End plus

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

source /etc/profile

Check whether the environment variable was set successfully

Verify Driven Version

cat /proc/driver/nvidia/version

Verify the cuda toolkit

nvcc -V

Test sample
Compile sample:

cd /usr/local/cuda-9.0/samples
sudo make

Waiting... (More than 10 minutes)

Compile binary files

cd bin/x86_64/linux/release
./deviceQuery

result = pass success

Finally, check the connection between the system and CUDA-Capable device:

./bandwidthTest

Install cudnn 7.1

Download the cloud disk address as above

Enter the file directory

command line

tar -xzvf cudnn-9.0-linux-x64-v7.1.tgz 
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 /usr/local/cuda/lib64/libcudnn*

Installation of Anaconda

After downloading, you need to install anaconda, enter the anaconda storage directory, and use the command

bash Anaconda3-4.2.0-Linux-x86_64.sh 

Keep in mind that sudo bash command is not allowed here

Tensorflow and keras installation

Use the command CONDA create-n tensorflow-gpu python = 3.5 to create a python 3.5 environment named tensorflow-gpu. Pay attention to writing the python version.  

Activation of conda environment

source activate tensorflow-gpu activates tensorflow-gpu environment

After activating the environment

pip install tensorflow-gpu==1.11

Test code

import tensorflow as tf
# Creates a graph.
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)
# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
# Runs the op.
print(sess.run(c))

Install Keras

pip install keras