Gnome crash after cuda 8.0 installation via dnf

asked 2017-08-04 03:15:00 -0600

ashr59 gravatar image

Yesterday, I decided to use PyTorch as my new deep learning framework.

After installation, I tried to run a program containing this flag:

CUDAVISIBLEDEVICES=0

But the program was not able to detect my GPU (nvidia Titan X) while I had no problem using my GPU with CUDA 7.5 in other frameworks (e.g. Caffe).

So, I tried to check the cuda activation in Python (anaconda) and realize that after "import torch" instruction, the answer to

torch.cuda.is_available() was FALSE.

After some investigations, I realized that the reason is maybe my Cuda version (7.5) So, I tried to install Cuda 8.0 by nvidia rpm file.

After cuda installation there was no improvement while the result of

nvcc --version was again Cuda 7.5.

So, I tried to install Cuda 8.0 directly from dnf. After that,

nvcc --version answer was Cuda 8.0 :)

But I still was unable to detect my GPU in Python (torch.cuda.is_available() was FALSE)

I thought that I should restart my system to finalize the changes.

After the restart, Gnome (I use Fedora 23) cannot be successfully loaded and my screen is blinking at " Starting Gnome Display Manager ..." boot step.

Any help will be appreciated.

edit retag flag offensive close merge delete