I recently upgraded from Fedora 41 to 43, and NVIDIA drivers went from version 570 to 580 (also tried 590, both open and proprietary). I started experiencing crashes with associated kernel oops that appear to be related to the PCIE alignment bug (noted when starting firefox and chromium). I manually downgraded to NVIDIA 570 (removed the rpm packages and used the NVIDIA installer) and this resolved the issue.
Has anyone else seen this? I’m using ASRock ALTRAD8UD-1L2T with RTX3060.
My next step will be applying PCIe alignment bug fixup patches to the fedora kernel and seeing if NVIDIA 580 and 590 will no longer cause crashes.
I was seeing the exact same thing in late Fedora 42. At that time I could work around it using older kernels. After upgrading to Fedora 43 there was no way to do this; the kernels wouldn’t match userland code.
I tried building a vanilla kernel with Ampere patches, but akmods wouldn’t work with that.
So I replaced my Nvidia card with a used AMD Radeon RX 6600. Very happy with this decision, because I can now have pre-boot graphics thanks to the AMD GOP driver. Somehow, this also finally made text ALT consoles work.
I might return to Nvidia when the nova driver is ready.
I’m using A3000 on my ALTRAD8UD-1L2T and see very frequent GPU failure errors (sometime my GPU just disappear that I have to restart the whole machine). Maybe I will try to downgrade to CUDA 12.8 (570) to see if the errors will disappear or not.
As anticipated, applying the pcie erratum patches to the kernel eliminates the crashes I was seeing with NVIDIA drivers newer than 570. It seems NVIDIA driver versions 575 and above trigger the Altra pcie bug. Unless you need CUDA, might as well go AMD now as others have mentioned, and avoid having to keep the NVIDIA drivers up to date along with managing your own patched kernel.
