Sorry if this is strictly a driver or X question and not a CUDA issue.
Every time I boot and want to log in, it gets stuck in the text mode of the boot-up messages, and then can’t activate the graphical screen. I always have to reboot into text mode (in grub enter “e” on the Linux line, then enter a “1” on the Linux line), and perform the recompile / reinstall with DKMS step. Then another reboot, and it works.
To me it seems that whenever I log into X (via LXDE) something is reconfigured that causes another login attempt to fail to start the graphical screen. The only fix I can see is the complete recompile/reinstall-with-DKMS procedure in text mode, and it gets pretty boring to have to do that every morning upon login.
Something destroys a working X configuration immediately after log-in. And if I don’t properly power down but crash the computer, same thing, so it must happen pretty early in the session.
Fedora 34, LXDE, tried with all Tesla drivers going back to 450.xx, all the way to the latest, 470.82.01. It doesn’t seem to be a driver thing, more like an X configuration thing. But in the end, I don’t know for certain, it’s beyond me.
Thanks for any help!
Update Nov 14:
It seems the kernel mod already expects that latest GTX/RTX driver, but I have to use the latest Tesla driver. I just upgraded to the latest Linux kernel, 5.14.17, and in the dkms kernel mod compilation / installation step I get these errors:
nvidia.ko.xz:
Running module version sanity check.
Error! Module version 470.82.01 for nvidia.ko.xz
is not newer than what is already found in kernel 5.14.17-201.fc34.x86_64 (495.29.05).
You may override by specifying --force
nvidia-modeset.ko.xz:
Running module version sanity check.
Error! Module version 470.82.01 for nvidia-modeset.ko.xz
is not newer than what is already found in kernel 5.14.17-201.fc34.x86_64 (495.29.05).
You may override by specifying --force.
nvidia-drm.ko.xz:
Running module version sanity check.
Error! Module version 470.82.01 for nvidia-drm.ko.xz
is not newer than what is already found in kernel 5.14.17-201.fc34.x86_64 (495.29.05).
You may override by specifying --force.
nvidia-peermem.ko.xz:
Running module version sanity check.
Error! Module version 470.82.01 for nvidia-peermem.ko.xz
is not newer than what is already found in kernel 5.14.17-201.fc34.x86_64 (495.29.05).
You may override by specifying --force.
Well, do the NVidia or Fedora people assume that everyone who wants the latest Linux kernel also wants the latest NVidia driver? Do they expect all hosts with Tesla cards to be headless (aka: without display) for remote-only login? The latest Tesla driver, 470.82.01 power my GTX980 just fine (even though GTX is not mentioned in the supported devices / release notes), but why does the 5.14.17 kernel upgrade expect to see the 495.29.05 NVidia driver? Why not take the one that is currently in use and expect the user wants that?
Where/how to I specify the --force suggestion? That must be some dkms or kernel mod compilation configuration option.