Just a heads up, many of the NVIDIA drivers are being bumped to 580 regardless of what package you have installed. This is the case for 20.04, 22.04, 24.04 and for driver versions 550 and 575 at least (and some others).
Say you are happy with your 550 drivers; then hold them in place:
sudo apt-mark hold nvidia-driver-550 libnvidia-common-550
If you've already upgraded - it might be too late to go back.
I'm not sure what NVIDIA's reasons are for forcing people to move on to new drivers - it doesn't seem to match up with how I understand LTS distributions to work.
I'm now trying to figure out how to get back to a set of 550 drivers - because some applications using the 580 drivers are freezing in situations that they didn't before.
If you have any hints, let me know.
The other complicating factor is that I'm on the 6.14 kernel at the moment. I may have to drop back to 6.11 as I think there's a version of the 550 driver packages still available to me that will build.
Thanks I wanted to avoid the .deb route and stick with Proprietary GPU Drivers : “Graphics Drivers” team - but I've hit a brick wall because the 550 (and I believe 575) drivers that are still on the PPA (e.g. 550.144.03-0ubuntu0~gpu24.04.1) don't build with kernel 6.14.
But yes, I've resigned myself to getting all the .deb's together and installing that way for now.
Once I've had a chance to test the .deb approach; I'll mark a solution.
NVIDIA drivers are, in my experience, quite unstable - and newer versions have regressions all the time.
Some games I've tested in wine now freeze when I switch (Alt + Tab) away from them, particularly in virtual desktop mode. I've tested a version of wine that I am positive was working just in case it was a regression that I missed there, but it would also freeze.
This doesn't happen every time so it's something racy in the NVIDIA driver.
Using kernel 6.11 (i.e. downgrade from HWE which is at 6.14) and the debian package available at version 550.144.03-0ubuntu0~gpu24.04.1 on ppa:graphics-drivers/ppa.
The first two I'm hoping may nudge the maintainers to reverse the decision and bring back 550.163.01. The last one is just in case there's something that can be done to mitigate the problem for the specific applications I'm having issues with.
That said, I've also noticed that vkgears falls over if I try to resize the window (e.g. __NV_PRIME_RENDER_OFFLOAD=1 vkgears followed by using mouse to drag the size). Not sure if this is a regression yet.
Thing with the orig tarballs is I'm pretty sure that they won't contain any Ubuntu-specific (or Debian-specific) tweaks; hence why they are a little lower on my list of things to try.
I’m still using 535 and all seems well but I’be been away for a while and have a lot up updates stacked up. I’m on 22.04 and looking at the update list it looks like it is “holding” 535 and updating them.
I should be OK, took quite an effort to get CUDA working on my RTX 3070 so I’d hate to break it now, 535 is not the “tested, recommended” version. Did my installing them using “Additional Drivers” tool automatically force a “hold”.
I’m out of my depth here, just asking to learn and understand. I broke 20.04 badly trying to get CUDA running on my laptop’s GTX 1060 (I believe). I now have it dedicated to running my AI security camera system so it has had updates turned off since I dedicated it to this usage a couple of years ago. Its only external network access is to send Email alerts when it detects someone on my property.
I will wait a bit to see if I get any responses here before doing these “reboot required” updates.
For now, I'm settling on the 550.144.03 line of drivers. I am still looking at how to get the 550.163.01 drivers from Debian Trixie installed.
Anyway, to remove the existing 550 and 580 drivers and install 550.144.03 and hold:
# remove the master packages
sudo apt remove nvidia-driver-550 nvidia-driver-580 --purge
# clean up the dependencies of the master (or meta) package
sudo apt autoremove --purge
# for some reason these were still installed; so removed them manually
sudo apt remove libnvidia-compute-580 libnvidia-compute-580:i386
# now install the 550.144.03 drivers
sudo apt install nvidia-driver-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-common-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-gl-550=550.144.03-0ubuntu0~gpu24.04.1 \
nvidia-dkms-550=550.144.03-0ubuntu0~gpu24.04.1 \
nvidia-kernel-common-550=550.144.03-0ubuntu0~gpu24.04.1 \
nvidia-kernel-source-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-extra-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-decode-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-encode-550=550.144.03-0ubuntu0~gpu24.04.1 \
nvidia-utils-550=550.144.03-0ubuntu0~gpu24.04.1 \
xserver-xorg-video-nvidia-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-cfg1-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-fbc1-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-decode-550:i386=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-encode-550:i386=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-fbc1-550:i386=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-gl-550:i386=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-compute-550=550.144.03-0ubuntu0~gpu24.04.1 \
nvidia-compute-utils-550=550.144.03-0ubuntu0~gpu24.04.1 \
libnvidia-compute-550:i386=550.144.03-0ubuntu0~gpu24.04.1
# hold to prevent updates; for some reason also need to hold libnvidia-common-550
sudo apt-mark hold nvidia-driver-550 libnvidia-common-550
On my 24.04.3, I am happy with the proprietary Nvidia driver v570. But when I tried to hold it as you described, I noticed that the “s” in the end is too much. It’s -driver-, not -drivers-:
s@Blacky-u16:~$ sudo apt-mark hold nvidia-drivers-570
E: Unable to locate package nvidia-drivers-570
E: No packages found
gs@Blacky-u16:~$ sudo apt-mark hold nvidia-driver-570
nvidia-driver-570 set on hold.
I wanted to avoid the .deb route and stick with