Getting the blasted nV drivers working again

[fivemack]

This is after a do-release-upgrade to 16.04

sudo apt-get update
sudo apt-get upgrade
sudo reboot now
sudo apt-get purge “cuda*” *nvidia*”
sudo dpkg -i /scratch/cuda-repo-ubuntu1604-8-0-local_8.0.44-1_amd64.deb 
sudo apt-get update
sudo apt-get install cuda

At this point 'nvidia-smi' works and says I'm on version 367 of the driver.

sudo apt-get autoremove
sudo /scratch/NVIDIA-Linux-x86_64-375.26.run

and I get an error about 'Unable to load the "nvidia-drm" kernel module.' but nvidia-smi works and says I'm on version 375 of the driver. However, CUDALucas says

device_number >=  device_count ... exiting
(This is probably a driver problem)

If I reboot and run /scratch/NVIDIA-Linux-x86_64-375.26.run again I get the same sequence of problems.

[chalsall]

Originally Posted by fivemack

If I reboot and run /scratch/NVIDIA-Linux-x86_64-375.26.run again I get the same sequence of problems.

Just to reflect...

I wanted to see what worked under Fedora Core MATE v25, and so installed everything involving CAD and 3D.

I found, to my surprise, that at least half of the programs grabbed the screens and didn't return them to the state they had taken them from (this is in a three screen configuration).

Sometimes I had to Ctnl-2 to get to a log-in screen, log in as root, and "shutdown -r now".

Other times I had to turn off the power switch, wait a couple of seconds, and then turn the power switch back on.

[Mark Rose]

Sounds like you're not using dkms to manage kernel modules. I bet you'll find the old 367 module under /lib somewhere.

[chris2be8]

Try:

sudo lspci -v | grep -i nvidia

And check it says the nvidia kernel driver is in use.

Chris

[bgbeuning]

I don't know why nvidia on Linux is so much more painful than on Windows but it is.

The last time I tried to use the "run" file, I found it did not cleanly uninstall the old driver.
Using "apt-get nvidia-375" worked better for me.