Could not load dynamic library ‘libcudnn.so.8‘；

install tensorflow When , We usually meet tf.test.is_gpu_available() Back to false The problem of , By this time GPU Is not available , And will prompt lib cudnn.so.8 non-existent , This is usually due to the following ：

1、 environment variable LD_LIBRARY_PATH Setting error ,

This variable should point to lib.cudnn.so.8 The parent directory of this file ;

resolvent ： Modify environment variables .

vi ~/.bashrc

Then modify the environment variable .

Just add the code with red box ;

2 、 No installation cudnn;

No installation cudnn Words , Want to go cuDNN Archive | NVIDIA Developer  Download the corresponding version , and cuda Version match , Otherwise, the installation will make an error , If it is linux Words , Download the deb file （deb The file should match its version ） Execute with the following command ：

sudo dpkg -i libcudnn8_8.1.1.33-1+cuda11.2_amd64.deb 

After installation, there will be libcudnn.so.8 This document , My path is /usr/lib/x86_64-linux-gnu Next , If it's not here , Can pass deb Under the document data.tar.xz Take a look below. libcudnn.so.8 Where is it installed .