..it’s always complicated. Basically, even if you get the official installer, something may not work at the end. I previously wrote about it on the post Cuda on CentOS 7. What you get at the end of the procedure described there, is the cuda libraries, but the drivers may not yet be fully operational. That is, you may be able to use the GPU to perform calculations, but not to display 3D graphics. To fix that, I found on advanced clustering technologies inc a good protocol that I will customize for my systems in the next lines.
1 – yum -y update, ./yum_install.sh (to install all our default packages)
2- Edit /etc/default/grub. Add “rd.driver.blacklist=nouveau nouveau.modeset=0″ to “GRUB_CMDLINE_LINUX”
3- grub2-mkconfig -o /boot/grub2/grub.cfg
4- reboot. After the reboot, you will have shitty graphics. Time to install the drivers.
5- CTRL+ALT+F2, log in as root, init 3. Now you don’t have graphics.
6- ./cuda_8.X.run, reboot, test that your card is detected by running nvidia-smi
That’s it. The most complicated thing to remember is obviously the grub part (points 2 and 3). I hope this is NOT helpful in the future, since when I’m done installing the next 16 GPU servers I’m expecting to have peace for a while. So that I can dedicate my time to what I really like.
Writing Programming. But of course that will never happen.