If using a binary install, upgrade your CuDNN library to match.

升级tensorflow_gpu-0.11 到tensorflow_gpu-0.12后，运行gpu相关程序，会出现如下error：

E tensorflow/stream_executor/cuda/cuda_dnn.cc:378] Loaded runtime CuDNN library: 5005 (compatibility version 5000) but source was compiled with 5105 (compatibility version 5100).  If using a binary install, upgrade your CuDNN library to match.  If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration.

意思是，当前的调用的cuDNN库是cuDNN-5.0， 但是tensorflow源是由cuDNN-5.1编译的，版本不匹配，因此需要升级cuDNN 到5.1

升级方法就是去NVidia官网下载 cuDNN-5.0 （我的是cuda-8.0支持的cuDNN5.0），然后按照官网方法复制或移动文件到/usr/local/cuda-8.0/ (我的文件存放结构)下

$ sudo mv include/cudnn.h /usr/local/cuda-8.0/include/
$ sudo mv lib/libcudnn*  /usr/local/cuda-8.0/lib64
$ sudo ln -s  /usr/local/cuda-8.0/lib64/libcudnn* /usr/local/cuda/lib64/

升级后，直接运行程序即可，不需要编译tensorflow-0.12源码