Now everything should be perfect with the nvidia driver (akmod-nvidia)!!
Xfce display compositing is a feature within the Xfwm4 window manager that manages window transparency, shadows, and special effects while reducing screen tearing. It acts as an intermediate layer between applications and the display, using GPU acceleration to stack windows and render visual effects like fades and transparency.It sucks. I hate it. That is all I need to know.
I began to think. If this was as serious a bug as it seems to be, why hasn't a fixed package come along? Maybe this is somehow specific to XFCE, as unlikely as that seems. (It is!!!) I do a search along those lines and hit paydirt immediately with "fedora xfce trouble with nvidia driver".
Fedora XFCE Nvidia issues (flickering, freezes, invisible windows) are commonly caused by Xorg/compositor conflicts or broken driver modules after updates.Fix Glitching/Invisible Windows: In XFCE, go to Settings -> Window Manager Tweaks -> Compositor and ensure the VSync is managed properly, or try setting Xfwm to xpresent to resolve flickering issues common with recent NVIDIA drivers.
This certainly describes what I am experiencing. There is indeed a "Window Manager Tweaks" settings thing with a Compositor tab, but I see nothing about VSync. Maybe this is because I am running nouveau right now and some menu item will appear when I try nvidia again.A suggestion is to do this:
xfwm4 --replace --vblank=glx & xfconf-query -c xfwm4 -p /general/vblank_mode -s glxThe first command tries it out, the second makes it permanent. When I just type "xfwm4", it tells me "another window manager (sfwm4) is running on screen :0.0". As near as I can tell this is telling me that xfwm4 is already running and I shouldn't have two instances of the same window manager running.XXXX NOTE: scroll to the bottom for details. This problem (on "trona") was fixed by doing the upgrade from 42 to 43 which put me back to running the nouveau driver.
On 1-11-2026 I did another long overdue "dnf update". I should figure out how to get the dnf auto thing to work properly (like it used to) and install updates as they come along. I got messages like this:
Autoinstall of module nvidia/575.57.08 for kernel 6.18.4-100.fc42.x86_64 (x8 >>> Building module(s).......................................(bad exit status: 2 >>> Failed command: >>> 'make' -j8 KERNEL_UNAME=6.18.4-100.fc42.x86_64 modulesI am still running an old 6.17.8 kernel.
A reboot will take me to 6.18.4.
Of course I worry that rebooting with a broken nvidia setup will get me into troubles. Something to try first might be:dnf remove akmod-nvidia dnf install akmod-nvidiaI do this and there are no ugly messages. Now I will try the reboot.It comes up as a useless disaster.
It at least has the correct resolution.
There are long delays after every character I type in a terminal windows. Windows turn white or disappear altogether, then come back..I reboot and select the previous 6.17.8 kernel -- it is the same train wreck.
I reboot again and select an even older 6.15.9 kernel.
This works beautifully!!What video driver am I using?
I try this:uname -a Linux trona 6.15.9-201.fc42.x86_64 #1 SMP PREEMPT_DYNAMIC Sat Aug 2 11:37:34 UTC 2025 x86_64 GNU/Linux lspci -k | grep -EA3 'VGA|3D|Display' 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) Subsystem: Gigabyte Technology Co., Ltd Device 4074 Kernel driver in use: nvidia Kernel modules: nouveau, nova_core, nvidia_drm, nvidiaI put the above into a script named 'vid'.
Then I reboot and run the latest. I use my "vid" command and see:Linux trona 6.18.4-100.fc42.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jan 8 18:43:17 UTC 2026 x86_64 GNU/Linux 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) Subsystem: Gigabyte Technology Co., Ltd Device 4074 Kernel driver in use: nvidia Kernel modules: nouveau, nvidia_drm, nvidiaI boot back to 6.15.9 and get a crazy reduced video resolution. This is useless in a yet different way. I try this with 6.18.4 and reboot.dnf remove akmod-nvidiaThis does not fix my problem.Just ugrade to F43
I do the usual dnf upgrade, and it comes up running OK! Note that I do not have akmod-nvidia unless it somehow got installed (it shouldn't) as part of the upgrade. In fact, I run "vid" and see:Linux trona 6.18.4-200.fc43.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jan 8 17:35:49 UTC 2026 x86_64 GNU/Linux 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) Subsystem: Gigabyte Technology Co., Ltd Device 4074 Kernel driver in use: nouveau Kernel modules: nouveauSo I am now running nouveau and it is doing pretty good so far.Switching to nouveau and back again
This problem looks like a buggy akmod-nvidia driver, so I am going to wait for a new version before installing it again. The current driver is 3:580.119.02-1.fc43. I have not yet installed this on F43. The kernel I am now running is 6.18.4-200.fc43.x86_64. A new akmod or a new kernel may merit trying it again. I could (should?) try dropping back to a previous version of the akmod, but I really don't want to invest more time in this right now.To go back to nouveau from nvidia is a matter of removing the nvidia packages and rebooting (or so they say).
dnf remove *nvidia* dnf install nvidia-gpu-firmwareSearching vi dnf.list shows lots of packages that mention nvidia. In particular:xorg-x11-drv-nvidiaCheck the file /etc/default/grub as it typically has nouveau in a "blacklist". This happens when you enable/install the nvidia drivers.
See this for lots of details:Installing older versions of akmod (or any package)
Find out what is available via:dnf list --showduplicates akmod-nvidia Available packages akmod-nvidia.x86_64 3:580.95.05-1.fc43 rpmfusion-nonfree akmod-nvidia.x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-updatesSo, 580.95 might be worth a try.There is a dnf downgrade command. The package must already be installed to do a downgrade. The general behavior is to go one version back, if there is such a version available. Or you can specify the exact version as:
dnf downgrade akmod-nvidia-3:580.95.05However, whenever you do an update, it will bump you to the latest version. There are two ways to avoid this. One is:dnf update --exclude=akmod-nvidiaThe other is to edit /etc/dnf/dnf.conf and make it look like this:[main] exclude=package_name_1,package_name_2In my case, just[main] exclude=akmod-nvidiaDownload the akmod-nvidia RPM
My idea is to avoid the DNF repository method and keep the RPM file(s) on my system and download exactly the version I want explicitly. But where do we download from? Their top level menu offers me 41, 42, and 43 (at this time), and either release or update packages for each. There is also "free" and "non-free". For akmod-nvidia you want non-free. I find:F41 - updates - akmod-nvidia-580.82.09-1.fc41.x86_64.rpm F42 - updates - akmod-nvidia-580.119.02-1.fc42.x86_64.rpm F43 - release - akmod-nvidia-580.95.05-1.fc43.x86_64.rpmI download these and place them in /Projects/akmodThere are a variety of other akmod packages, as well as kmod packages, and I have no idea how to sort all these out. Perhaps the HOWTO link above would explain.
I do this:
dnf install akmod-nvidia-580.95.05-1.fc43.x86_64.rpm Updating and loading repositories: Repositories loaded. Package Arch Version Repository Size Installing: akmod-nvidia x86_64 3:580.95.05-1.fc43 @commandline 96.6 KiB Installing dependencies: akmods noarch 0.6.2-3.fc43 updates 65.3 KiB fakeroot x86_64 1.37.1-3.fc43 fedora 150.8 KiB fakeroot-libs x86_64 1.37.1-3.fc43 fedora 124.6 KiB kmodtool noarch 1.1-14.fc43 fedora 27.6 KiB python3-progressbar2 noarch 4.5.0-5.fc43 fedora 609.9 KiB python3-rpmautospec-core noarch 0.1.5-8.fc43 fedora 11.3 KiB python3-utils noarch 3.9.1-2.fc43 fedora 243.2 KiB rpmdevtools noarch 9.6-13.fc43 fedora 213.8 KiB xorg-x11-drv-nvidia-kmodsrc x86_64 3:580.95.05-1.fc43 rpmfusion-nonfree 86.7 MiB Installing weak dependencies: python3-rpmautospec noarch 0.8.3-2.fc43 fedora 453.2 KiB Transaction Summary: Installing: 11 packagesNow we reboot and see what happens. Well it tells me it is rebuilding /usr/src/akmods/nvidia-kmod.latest (which is 580.95), but then it tells me it cannot load nvidia and is falling back to nouveau. And it comes up in proper 3840x2160 resolution, but indeed running nouveau:Linux trona 6.18.4-200.fc43.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jan 8 17:35:49 UTC 2026 x86_64 GNU/Linux 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) Subsystem: Gigabyte Technology Co., Ltd Device 4074 Kernel driver in use: nouveau Kernel modules: nouveauNow, for the heck of it I do a "dnf update". I get a new kernel (6.17.8) and it updates akmod-nvidia:Upgrading: akmod-nvidia x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-u 96.8 KiB replacing akmod-nvidia x86_64 3:580.95.05-1.fc43 @commandline 96.6 KiBI thought doing an explicit "dnf install xx.rpm" would avoid this, but clearly not. I go ahead with this update and after a long wait get a message:Autoinstall of module nvidia/575.57.08 for kernel 6.18.5-200.fc43.x86_64 (x86_64) >>> Building module(s).....................................(bad exit status: 2) >>> Failed command: >>> 'make' -j8 KERNEL_UNAME=6.18.5-200.fc43.x86_64 modules >>> >>> Error! Bad return status for module build on kernel: 6.18.5-200.fc43.x86_64 (x86_64) >>> Consult /var/lib/dkms/nvidia/575.57.08/build/make.log for more information. >>> >>> Autoinstall on 6.18.5-200.fc43.x86_64 failed for module(s) nvidia(10).I am still running the 6.18.4 kernel, so perhaps this will clean up when I reboot to the new 6.18.5. I go ahead and reboot.I am running nvidia now, but it is buggy and useless. Nouveau is bad, but the current state of nvidia is worse. How to revert?
su dnf remove akmod-nvidia rebootThis does not do it! It boot up still running nvidia!
Let's try something more drastic:dnf remove *nvidia* dnf install nvidia-gpu-firmwareThis gets rid of a lot of packages. Maybe after this "cleanup" I can install akmod-nvidia with success?dnf remove *nvidia* Package Arch Version Repository Size Removing: kmod-nvidia-6.18.4-100.fc42.x86_64 x86_64 3:580.119.02-1.fc42 @commandline 35.4 MiB kmod-nvidia-6.18.4-200.fc43.x86_64 x86_64 3:580.119.02-1.fc43 @commandline 35.4 MiB kmod-nvidia-6.18.5-200.fc43.x86_64 x86_64 3:580.119.02-1.fc43 @commandline 35.4 MiB kmod-nvidia-open-dkms noarch 3:575.57.08-1.fc41 cuda-fedora41-12-9-local 102.4 MiB nvidia-gpu-firmware noarch 20260110-1.fc43 updates 101.0 MiB nvidia-kmod-common noarch 3:575.57.08-1.fc41 cuda-fedora41-12-9-local 99.2 MiB nvidia-modprobe x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 50.9 KiB nvidia-settings x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 4.4 MiB xorg-x11-drv-nvidia x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 169.2 MiB xorg-x11-drv-nvidia-cuda-libs x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 345.6 MiB xorg-x11-drv-nvidia-libs x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 443.2 MiB xorg-x11-drv-nvidia-power x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 2.3 MiB xorg-x11-drv-nvidia-xorg-libs x86_64 3:580.119.02-1.fc43 rpmfusion-nonfree-update 19.4 MiB Removing unused dependencies: cuda-opencl-12-9 x86_64 12.9.19-1 cuda-fedora41-12-9-local 91.7 KiB cuda-toolkit-12-9-config-common noarch 12.9.79-1 cuda-fedora41-12-9-local 0.0 B cuda-toolkit-12-config-common noarch 12.9.79-1 cuda-fedora41-12-9-local 44.0 B cuda-toolkit-config-common noarch 12.9.79-1 cuda-fedora41-12-9-local 41.0 B dkms noarch 3.3.0-1.fc43 updates 217.0 KiB egl-gbm x86_64 2:1.1.2.1-2.fc43 fedora 29.3 KiB egl-wayland x86_64 1.1.21-1.fc43 updates 83.4 KiB egl-x11 x86_64 1.0.4-1.fc43 updates 165.8 KiB gcc-c++ x86_64 15.2.1-5.fc43 updates 41.4 MiB kernel-devel-matched x86_64 6.18.5-200.fc43 updates 0.0 B opencl-filesystem noarch 1.0-23.fc43 fedora 0.0 B Transaction Summary: Removing: 24 packages dnf install nvidia-gpu-firmwareAfter this, I reboot and am running nouveau. It has some minor issues, but at this time nvidia is entirely useless. Screens go white and/or disappear entirely. The crazy activity is triggered by both mouse and keyboard events. I've had enough of this for today. I used to run the nvidia driver with complete satisfaction. I wonder what went wrong?Other that using my "vid" script, I good way to detect if nouveau is running is to put the cursor in the window running my python temp_plotter script. If motion is smooth, you have nvidia. If it is choppy and all but useless, you are running nouveau.
Try again a week later
I got tired of fooling with this, so I just ran with nouveau for a week. The experience is pretty bad. Mouse motion is choppy and unpredictable. Such bad performance is ridiculous and inexcusable with this video card:lspci | grep VGA 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) vid Linux trona 6.18.5-200.fc43.x86_64 #1 SMP PREEMPT_DYNAMIC Sun Jan 11 17:09:32 UTC 2026 x86_64 GNU/Linux 04:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1) Subsystem: Gigabyte Technology Co., Ltd Device 4074 Kernel driver in use: nouveau Kernel modules: nouveauSo, I have decided to try again. My hope is that the big cleanout of nvidia packages above may have gotten rid of my problem. And I suspect my problem is related to a tangle of packages I installed when I decided to try CUDA programming a few months ago.
There is valuable information in this "HOWTO". It says that you can revert to running nouveau without uninstalling packages by doing the following:With recent drivers as packaged with RPM Fusion, it is possible to switch easily between nouveau and nvidia while keeping the nvidia driver installed. When you are about to select the kernel at the grub menu step. You can edit the kernel entry, find the linux boot command line and manually remove the following options "rd.driver.blacklist=nouveau modprobe.blacklist=nouveau". This will allow you to boot using the nouveau driver instead of the nvidia binary driver. At this time, there is no way to make the switch at runtime.I do this:su dnf update dnf install akmod-nvidia Installing: akmod-nvidia x86_64 3:580.119.02-1.fc43 rpmfusio 96.8 KiB Installing dependencies: akmods noarch 0.6.2-3.fc43 updates 65.3 KiB egl-gbm x86_64 2:1.1.3-1.fc43 updates 29.3 KiB egl-wayland x86_64 1.1.21-1.fc43 updates 83.4 KiB egl-x11 x86_64 1.0.4-1.fc43 updates 165.8 KiB fakeroot x86_64 1.37.1-3.fc43 fedora 150.8 KiB fakeroot-libs x86_64 1.37.1-3.fc43 fedora 124.6 KiB kernel-devel-matched x86_64 6.18.5-200.fc43 updates 0.0 B kmodtool noarch 1.1-14.fc43 fedora 27.6 KiB nvidia-modprobe x86_64 3:580.119.02-1.fc43 rpmfusio 50.9 KiB nvidia-settings x86_64 3:580.119.02-1.fc43 rpmfusio 4.4 MiB opencl-filesystem noarch 1.0-23.fc43 fedora 0.0 B python3-progressbar2 noarch 4.5.0-5.fc43 fedora 609.9 KiB python3-rpmautospec-core noarch 0.1.5-8.fc43 fedora 11.3 KiB python3-utils noarch 3.9.1-2.fc43 fedora 243.2 KiB rpmdevtools noarch 9.6-13.fc43 fedora 213.8 KiB xorg-x11-drv-nvidia x86_64 3:580.119.02-1.fc43 rpmfusio 169.2 MiB xorg-x11-drv-nvidia-cuda-libs x86_64 3:580.119.02-1.fc43 rpmfusio 345.6 MiB xorg-x11-drv-nvidia-kmodsrc x86_64 3:580.119.02-1.fc43 rpmfusio 87.4 MiB xorg-x11-drv-nvidia-libs x86_64 3:580.119.02-1.fc43 rpmfusio 443.2 MiB xorg-x11-drv-nvidia-xorg-libs x86_64 3:580.119.02-1.fc43 rpmfusio 19.4 MiB Installing weak dependencies: python3-rpmautospec noarch 0.8.3-2.fc43 fedora 453.2 KiB xorg-x11-drv-nvidia-power x86_64 3:580.119.02-1.fc43 rpmfusio 2.3 MiB Transaction Summary: Installing: 23 packagesAnd then of course, I reboot to try it out.
It is a miserable failure. Nouveau is bad, but this is much worse. The problem is that windows blank out on almost every keystroke, then come back and go away at random.I edit the grub command line to get back to nouveau. This takes several tries to get right. You type "e" to edit the current boot line, then use arrow keys and the DEL key to change it. I use F10 when done to boot my system.
I see "rhgb" in the command line -- something I hate and would like to get rid of. This gets me back to running nouveau, which is what I am using as I type this.
The grub line now mentions something called "nova_core" and has it on the blacklist along with nouveau, something like this:
modprobe.blacklist=nouveau,nova_core modprobe.blacklist=nova_coreI leave it blacklisted as shown, not knowing what it is. I search and learn:nova_core is an emerging, experimental open-source Rust-based kernel driver for modern NVIDIA GPUs (GSP-based), designed by Red Hat engineers to eventually replace the older Nouveau driver.It must be blacklisted to use the nvidia driver, so I probably did well to leave it blacklisted at any event.Continued in a subsequent page.
Have any comments? Questions? Drop me a line!Adventures in Computing / tom@mmto.org