I’m trying to debug a problem where my system crashes if I either set-default to graphical or if I use “isolate” to go to graphical mode. As a test I’d like to start up a window session manually after logging in on the terminal in multi-user. There’s no startx and no xinit, and even the AI answers from Google have been useless. What’s the newer “better” way to do this?
Hi,
A google search found this https://forums.fedoraforum.org/showthread.php?335338-Fedora-43-Wayland-does-not-start-directly
Which suggests gnome-shell --wayland should work.
Regards Tom
Thanks. That did the trick.**
[**Where the trick is to instantly crash my system. That means this is almost certainly a GPU/driver issue, and not an issue with the systemd initialization process or associated privs. Which is weird because we have other identical systems (Dell Prec 5860 with T1000 GPU), that have the identical install (ansible-controlled), and no issues. Which would make me think gpu card problem, except that I forgot to remove Windows, and that still works fine. And I ran every BIOS diagnostic I could and they found no problems.]
Hi,
If you’re dell precision is anything like mine, it should support hybrid mode. So you can at least start gnome with the intel chip. There should be an option to enable both in the bios. I’m currently running hybrid to save on battery consumption and just use nvidia when needed, (for example: __NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia steam ).
Are you using the drivers from the nvidia repo?
If so an update may updated to graphics drivers not support by your gpu. The repo if enabled is set to automatically installed the latest driver, but these are for newer cards.
So the driver that should be installed is 580.178.04-1.el10
Even if you not using the official repo, I would remove the current drivers installed and follow the guide provided here:-
sudo dnf install cuda-drivers-580 -y
Should get you on the latest drivers, are more precise installation would be:-
sudo dnf install \
--refresh \
--repo=cuda-rhel10-x86_64 \
'nvidia-driver-3:580.178.04-1.el10.x86_64' \
'cuda-drivers-3:580.178.04-1.el10.x86_64' \
'kmod-nvidia-latest-dkms-3:580.178.04-1.el10.x86_64'
You can even use the dnf version lock as advised or just disable the repo.
The 580 Drivers don’t get updated that often.
The last release was back in July.
Regards Tom.
After a failed GUI session, you can log in using command line and look at the earlier logs to see why the GUI session went wrong.
Whatever is causing the crash is happening so fast that it leaves no useful logs behind. I even tried “sync; sync; sync; systemctl isolate graphical.target; sync; sync; sync; for i in {1..10000}; do sync;” but nothing useful was revealed. So it has to be either entirely hardware (but Windows works) or a hardware/software interaction from Linux. But we have a dozen other identical systems with the same package set installed that have no issues.