TensorFlow CUDNN Common CUDNN Error Resolution

When you execute TensorFlow code with Spyder, "Kernel died,restarting" is displayed each time. Find out why the terminal window has the following error:Loaded Runtime cudnn library:5005 (Compatibility version 5000) but source is compiled with 5103 (compatibility vers Ion 5100). 1. According to the prompts, should be when the CUDNN version of the problem, itself installed when CUDA7.5, if the original installation CUDNN, you need to first copy several files deleted: CD/USR/LOCAL/CUDA/CD include Sud       o rm cudnn.h cd lib64 sudo rm libcudnn* You do not need to perform this step if CUDNN has not been installed.       2. Then download the latest version of the CUDNN, pay attention to the latest version, I used the cuda7.5, first installed CUDNN v5.0, found or not, and then installed the v5.1 on the good. Login to nvidia website to download: Https://developer.nvidia.com/rdp/cudnn-download The latest version is: Cudnn v5.1 Library for Linux
3. After downloading, use the following command to extract and copy to Cuda OK: tar-zxf cudnn-7.5-linux-x64-v5.1-prod.tgz cd cuda sudo cp lib64/*/usr/l Ocal/cuda/lib64/sudo CP include/cudnn.h/usr/local/cuda/include/