Tensorflow GPU installation process record

1. Install first cuda. download cuda_8.0.44_linux.run, install , The installation path /usr/local/cuda-8.0
2. install cudnn. download cudnn-8.0-linux-x64-v6.0.tgz, Extract the file to /home/weizhong/software/cuda, hold /home/weizhong/software/cuda/includ In the catalog cudnn.h Copied to the /usr/local/cuda-8.0/include
3. install anaconda. download Anaconda3-5.0.1-Linux-x86_64.s4. establish conda Environmental Science . conda create -n tensorflow python=3.5.  Here because tensorflow 1.4 Version only supports python3.5
5. Uninstall the existing tensorflow： pip uninstall tensorflow-gpu
   pip uninstall protobuf
6. source activate tensorflow, Here it is conda In the environment , sudo pip3 install tensorflow-gpu==1.4
notes ：
1. When I write my blog ,cuda The latest version is 9.1,tensorflow The latest version of is 1.5, But in my 12 Lenovo computer manufactured in （NVIDIA GF630M) On tensorflow1.5 Yes cuda9.1 I won't support it , After unsuccessful installation for many times , I have to lower cuda Version to 8.0, Corresponding tensorflow Version down to 1.4. Then the installation is successful .
2. install cuda when , There is an important question , If it shows
“You are attempting to install on an unsupported configuration.”
Such information , In the following
Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 xxx xx?
In the options , Input “n” Means not to install , Otherwise, the installation will not succeed , I repeat here many times , Last in autocyz With the help of your blog post . thank autocyz, See ：ubuntu16.04 install cuda7.5 （ website ：http://blog.csdn.net/autocyz/article/details/51841157）
3. install cuda and cudnn To add them to the environment variables , Available for Du Niang . I'm referring to gtx_viewer 's blog post “Nvidia CUDA Toolkit Installation （cuda）”（ see ：https://www.cnblogs.com/jinggege/p/5766146.html）
sudo gedit /etc/profile
then , Add... At the end of the open file （ here cuda-8.0 Is my cuda Installation path for ,/home/weizhong/software/cuda Express cudnn Decompression path )：
export PATH=/usr/local/cuda-8.0/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/home/weizhong/software/cuda/lib64:$LD_LIBRARY_PATH
Add again lib The library path ：
sudo vim /etc/ld.so.conf.d/cuda.conf
The contents are as follows ：
/usr/local/cuda-8.0/lib64
/home/weizhong/software/cuda/lib64
/cudnn_dir/lib64
/lib
/usr/lib
/usr/lib32
4. establish conda The environment is very important , It can save a lot of trouble , Then install in the created environment tensorflow