on machine that had a working 390.48 driver which stopped working when a kernel upgrade occurred, the 410.78 driver was incorrectly installed on CentOS 7.5 (kernel 3.10.0-957.el7.x86_64) and when we try to uninstall it using the NVIDIA-Linux-x86_64-410.78.run installer file, it appears to successfully uninstall the driver. We then install the correct driver file (NVIDIA-Linux-x86_64-390.87.run), which appears to install correctly (nvidia-smi displays correct driver info, correct version in the /proc/driver/nvidia/version file and modprobe nvidia also). We put the runlevel back to graphical and reboot. The machine fails to boot into X window environment and we have to log on through a terminal screen. When we try nvidia-smi, we get a version mismatch error mssg and see Nvidia 410.78 listed in version file but still see 390.87 with modprobe nvidia. It appears that the prior install (410.78) info is being retained in the system and then overwriting the 390.87 install. Any ideas on where to look for the 410.78 parameters being retained?
See if rebuilding the initrd using
helps with that. Otherwise, please run nvidia-bug-report.sh as root and attach the resulting .gz file to your post. Hovering the mouse over an existing post of yours will reveal a paperclip icon.
First, I would try “dracut -f” as @generix suggested.
If that doesn’t work, I would suggest the following:
As @levettrm420 mentioned the 410.78 install failed, you may find some remnants (like a build log file) that may give you some hints in the /var/lib/dkms/nvidia/410.78/build/ directory – that is, if such a directory still exists (it seems to get deleted in successful installs).
I just posted in a separate thread (about 410.78 on fc28 – which has similarities with CentOS 7.5) that I believe there to be a problem (concerning DRI2 and ACPI) with the 410.xx driver version that did not exist in 390.xx. So, having to fall back on the 390.87 driver as an interim solution seems to be both sensible and sane.
FWIW, I got my fc29 system running again (with 390.87) thanks to https://www.if-not-true-then-false.com/2018/fedora-28-workstation-install-guide/ . The “How to Patch nVidia Installer” section on that page makes reference to http://www.murga-linux.com/puppy/viewtopic.php?t=72405&start=75, which, in peebee’s post dated Sat 02 Apr 2016, 13:05, gives further hints for how to do the actual patch of the nvidia installer for the kernel version you may be using – but hopefully you will not have to go that deep.