tensorflow-gpu版本安装巨坑

1. 按照如下的教程基本可以调通Tensorflow-gpu 1.13.1+Python 3.7.2+CUDA 10.0 +cuDNN7.5的配置，简单的测试可以通过。

https://blog.csdn.net/weixin_44095128/article/details/88664928

2. 但是一旦GPU调用卷积模块CUDNN就报错如下：

2019-05-23 10:06:53.489842: E tensorflow/stream_executor/cuda/cuda_dnn.cc:324] Loaded runtime CuDNN library: 7.1.4 but source was compiled with: 7.4.1.  CuDNN library major and minor version needs to match or have higher minor version in case of CuDNN 7.0 or later version. If using a binary install, upgrade your CuDNN library.  If building from sources, make sure the library loaded at runtime is compatible with the version specified during compile configuration.

至于原因：

http://www.pianshen.com/article/5464281116/

已经给出，就是在使用Anaconda搭建python环境时会有一个默认的cudn和cuda版本对应问题，这个最好按照anaconda包里的版本去安装对应的cudn和cuda。

比如我的：

在路径：D:\Anaconda3\pkgs下有：cudnn-7.1.4-cuda9.0_0文件夹，就去安装下载对应的即可。

https://developer.nvidia.com/rdp/cudnn-archive

https://developer.nvidia.com/cuda-90-download-archive?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exelocal

并且与下链接中的版本对应:

https://www.cnblogs.com/tuji-sjp/p/10015634.html

注意:安装CUDA时选择简易安装，会自动更新相应的驱动版本，避免不必要的错误。

3.最终我的版本是

cuda_9.0.176_win10

cudnn-9.0-windows7-x64-v7.1

python3.5.6

anaconda下文件：cudnn-7.1.4-cuda9.0_0

tensorflow-gpu-1.8.0