Problem with cudnn
nagi1999a opened this issue · comments
I am using the image tag all-py36-cu111 to create LSTM model with Keras. However, I find that it will cause the following error:
Fail to find the dnn implementation.
And cause the following error in the console:
Loaded runtime CuDNN library: 8.0.5 but source was compiled with: 8.1.0
From the base image nvidia/cuda:11.1-cudnn8-devel-ubuntu18.04, I find that the cudnn version 8.0.5.39 is too old for the latest tensorflow-gpu package. Due to this, the base image maybe need to change to 11.2.2-cudnn8-devel-ubuntu18.04?
