ubuntu14.04安装tensorflow-gpu

ubuntu14.04 安装anaconda+tensorflow-gpu版本

本电脑为cuda-8.0，cudnn-6.0.21，安装tensorflow-gpu=1.3.0

更多匹配版本见  https://www.tensorflow.org/install/source#common_installation_problems

删除环境：conda env remove -n 环境名字

走过的歪路：

cuda-8.0，cudnn-5.1，安装tensorflow-gpu=1.2.1时tensorboard只显示graph，不显示loss值的图

更新cudnn后，为cuda-8.0，cudnn-6.0.21，安装tensorflow-gpu=1.4.1

出现错误：Loaded runtime CuDNN library: 7103 (compatibility version 7100) but source was compiled with 7005 (compatibility version 7000).

安装tensorflow-gpu=1.3.0,完美解决

错误：  No module named 'skimage'     解决方法： conda install scikit-image

 

• according to python official doc :

On Linux, Mac OS X and other POSIX systems, use the versioned Python commands in combination with the -m switch to run the appropriate copy of pip:

python2   -m pip install SomePackage  # default Python 2
python2.7 -m pip install SomePackage  # specifically Python 2.7
python3   -m pip install SomePackage  # default Python 3
python3.4 -m pip install SomePackage  # specifically Python 3.4

 

• 更改cudnn 

 https://blog.csdn.net/l297969586/article/details/67632608#commentBox

1、解压出一个名为cuda的文件夹，文件夹中有include和lib64两个文件夹 
2、删除原来的cudnn

sudo rm -rf /usr/local/cuda/include/cudnn.h
sudo rm -rf /usr/local/cuda/lib64/libcudnn*

3、安装安装需要版本的cudnn，在终端cd到刚解压的cuda文件夹

sudo cp include/cudnn.h /usr/local/cuda/include/
sudo cp lib64/lib* /usr/local/cuda/lib64/

4、cd到/usr/local/cuda/lib64/文件夹下，建立软链接（注意版本号换成你自己的）

sudo chmod +r libcudnn.so.5.0.5  
sudo ln -sf libcudnn.so.5.0.5 libcudnn.so.5  
sudo ln -sf libcudnn.so.5 libcudnn.so  
sudo ldconfig  

除了博客中的这4步还需加下面两行代码 

sudo chmod a+r /usr/local/cuda/include/cudnn.h 

sudo chmod a+r /usr/local/cuda/lib64/libcudnn*