CUDNN_STATUS_INTERNAL_ERROR after upgarde to tensorflow-gpu 1.9

Question

I am on Ubuntu 16.04 and am using python 2.7 and tensorflow-gpu 1.6 which works fine.

After upgrading to tensorflow-gpu 1.9 with pip I get the following error message:

could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR

I have CUDA 9.0 and cuDNN 7.0.5.

What am I missing here?

Answer 1

The solution given in RuntimeError: CUDNN_STATUS_INTERNAL_ERROR also worked for me. In short, run

sudo rm -rf ~/.nv

and reboot.