I had just tried installing cuda on my laptop (skylake cpu, 32GB mem, 1TB ssd, Nvidia 1200M graphics), running Fedora 27, but it appears that the cuda installing conflicts with the nvidia graphic installation.
Now I am thinking I may be better of on my workstation (i7 cpu, 32GB mem, 6TB disk, Nvidia GTX 950), but I’m hoping for a bit of guidance about using a graphics card for both graphics and GPU computation, specifically with Tensorflow.
I did install…
- the 9.2 cuda repo (cuda-repo-fedora27-9-2-local-9.2.88-1.x86_64.rpm) and then the packages within,
- cuDNN SDK v7
- graphics driver: xorg-x11-drv-nvidia-396.26-1.fc27.x86_64.rpm
But note that the graphics driver is 396.26 and yet the cuda repo contains several rpms of xorg-x11-drv-nvidia that are version 396.24. That appeared to be the basis of my problems. Is there a solution for this? Such as not installing the graphics drivers from the cuda repo?