[SOLVED] Broken Nvidia Driver - solved by updating to kernel 5.11

[klrcommute]

Suspect a recent update broke my Nvidia driver somehow. Was on kernel 5.4, had previously installed (long time ago) proprietary Nvidia driver with some instruction about switching from Nouveau. Everything was working just fine for many months until just recently.

Driver Manager now informs me there's no hardware acceleration on startup. My choices are nvidia-driver 460 (recommended)--was using this, definitely broken. nvidia-driver-465. Tried switching to this, still broken. xserver-xorg-video-nouveau. I get the following error when I try to switch back to this:

pk-client-error-quark: The following packages have unmet dependencies:
libllvm10: Depends: libatomic1 (>= 4. but it is not going to be installed
Depends: libedit2 (>= 2.11-20080614-0) but it is not going to be installed
Depends: libffi7 (>= 3.3~20180313) but it is not going to be installed
Depends: libstdc++6 (>= 9) but it is not going to be installed
(268)

I suspect this might be from tinkering back when I initially installed the proprietary nvidia driver. Purposefully disabled? Can't recall.

And lastly nvidia-driver-390. I haven't tried this one yet, but will do so shortly.

Just thought I'd run this by y'all as googling around for methods to repair has prettymuch exceeded what little terminal kung-fu I possess. Any suggestions? Thanks!

I also tried downloading 460.84 directly from nvidia, but I'm having trouble getting the file to run as root and have just gotten an error message to that effect from the installer so far.

[klrcommute]

Now things seem to be entirely broken (instead of just no hardware acceleration and wrong resolution). Just some colored lines/artifacts on boot. Seem to remember something similar in transitioning from nouveau to proprietary nvidia driver. Typing this from a different mint install (mate) using 384.130.

EDIT: Got the artifact screen fixed from recovery mode. Now I'm just back to my original problem with the broken Nvidia drivers and no hardware acceleration.

[SMG]

Driver Manager now informs me there's no hardware acceleration on startup.

Driver Manager does not give that warning. The Cinnamon desktop gives that warning because it detects that software rendering is being used.

(instead of just no hardware acceleration and wrong resolution).

That sounds like an issue with the Nvidia driver being slow to load. That seems to be a more common problem with the Nvidia-460 and newer drivers. Try restarting X Server with Ctl-Alt-Backspace. If that corrects the issue, a more permanent solution is in this post [SOLVED] NVIDIA driver failing sometimes on boot ("no hardware acceleration"). Let us know if the fix roblm recommends in the second post addresses the issue for you.

If restarting the X Server does not fix the issue, please give us information about your install by entering this command in a terminal: inxi -Fxxxrz
Click </> from the mini toolbar above the textbox where you type your reply and then place your cursor between the code markers and paste the results of the command between the code markers [code]Results[/code]. This will let us know how Mint sees your hardware.

[karlchen]

<mod> deeb's post about "a similar issue" has been turned into a sparate thread: LM 20 Cinnamon Broken Nvidia Driver </mod>

[klrcommute]

Just getting round to checking this again. Thanks for the reply. Will try what you recommend and look thru all the numerous nvidia driver issues threads that have since been posted.

[klrcommute]

Driver Manager now informs me there's no hardware acceleration on startup.

Driver Manager does not give that warning. The Cinnamon desktop gives that warning because it detects that software rendering is being used.

(instead of just no hardware acceleration and wrong resolution).

That sounds like an issue with the Nvidia driver being slow to load. That seems to be a more common problem with the Nvidia-460 and newer drivers. Try restarting X Server with Ctl-Alt-Backspace. If that corrects the issue, a more permanent solution is in this post [SOLVED] NVIDIA driver failing sometimes on boot ("no hardware acceleration"). Let us know if the fix roblm recommends in the second post addresses the issue for you.

If restarting the X Server does not fix the issue, please give us information about your install by entering this command in a terminal: inxi -Fxxxrz
Click </> from the mini toolbar above the textbox where you type your reply and then place your cursor between the code markers and paste the results of the command between the code markers [code]Results[/code]. This will let us know how Mint sees your hardware.

I take your point on the first quote, but at my level that's kind of a distinction without a difference. Boot up computer, that's funny, why's my resolution so low, oh look a little window that says driver manager says your stuff's all messed up.

Ok, so I'll explain this as best I can with my limited computer skills. Your Ctrl-Alt-Backspace suggestion to restart the X Server totally worked to set things right temporarily--upon using it everything reverts back to the proper higher resolution, presumably with hardware acceleration as I've lost my little cinnamon desktop driver manager warning.

The permanent workaround solution you suggest with

"nvidia
nvidia-drm
nvidia-modeset"

is actually something I've had implemented for a long time, so this was already in play before the issue began. I don't know, but I suspect maybe this caused me some of the problems with simply trying to switch back to the open-source xserver-xorg-video-nouveau driver? Because I never bothered to remove the nvidia-modeset code? No idea.

Back to what happened: Ctrl-Alt-Backspace did set things right, but upon restart they'd just revert back to being messed up. Things I tried since: Upgrading to nvidia 470.57.02 drivers. Gave me a bunch of grief about some package dependencies but installed nonetheless. Didn't fix anything. Upgraded to 20.2 Uma while I was updating everything in update manager to see if there anyone had fixed anything. Upgrading to 20.2 didn't seem to help either. While I was at it I switched from one of the last 5.8 kernels to the latest 5.11.0-25. This actually seemed to fix my problem, and mint appears to be reliably booting using the nvidia driver at the proper resolution with hardware acceleration.

Don't ask me what was going on there, or what got fixed--that's about the best I can explain it. I was going to grab you the inxi -Fxxxrz info while things were still broken, but fixed them by accident by installing the latest 5.11 kernel. (I suppose I could still do so if I revert back to 5.8 kernel to see if things break again. Let me know and I'll try it. I assume you're a developer?)

Anyway, just wanted to sum it up to say thanks for the help!

[SMG]

Your Ctrl-Alt-Backspace suggestion to restart the X Server totally worked to set things right temporarily--upon using it everything reverts back to the proper higher resolution, presumably with hardware acceleration as I've lost my little cinnamon desktop driver manager warning.

X Server runs in the graphics in Mint. That server starts during the boot cycle. Nvidia driver modules also get prepared during the boot cycle. However, there is nothing to tell X Server to wait if Nvidia is slow getting its act together to be ready. X Server starts and it loads what is ready. Eventually Nvidia gets ready. When you restart X Server manually, Nvidia is now ready and it gets priority.

The more permanent solution helps get Nvidia moving sooner in the boot cycle so it is ready when X Server starts the first time.

The permanent workaround solution you suggest with

"nvidia
nvidia-drm
nvidia-modeset"

is actually something I've had implemented for a long time, so this was already in play before the issue began. I don't know, but I suspect maybe this caused me some of the problems with simply trying to switch back to the open-source xserver-xorg-video-nouveau driver? Because I never bothered to remove the nvidia-modeset code?

That sounds like a plausible explanation. I would think the system would give errors, or at least be slowed down, if it could not find and load those modules.

Back to what happened: Ctrl-Alt-Backspace did set things right, but upon restart they'd just revert back to being messed up.

That would probably make sense to you now since I explained above what happens.

Things I tried since: Upgrading to nvidia 470.57.02 drivers. Gave me a bunch of grief about some package dependencies but installed nonetheless. Didn't fix anything. Upgraded to 20.2 Uma while I was updating everything in update manager to see if there anyone had fixed anything. Upgrading to 20.2 didn't seem to help either. While I was at it I switched from one of the last 5.8 kernels to the latest 5.11.0-25. This actually seemed to fix my problem, and mint appears to be reliably booting using the nvidia driver at the proper resolution with hardware acceleration.

When you install an Nvidia driver, that just installs files to your computer. They can not do anything at that point.

After installing the files, you then need to reboot the computer because the boot process builds and loads the Nvidia modules. Those modules will only work with the kernel you have installed which means each time you install a new (or updated) kernel, the modules need rebuilt in order to work. It is possible that just rebuilding the modules is the reason it works. Or this driver version may work better with the 5.11 kernel than the 5.8 kernel. I do not know, but I also do not argue with what is working.

Don't ask me what was going on there, or what got fixed--that's about the best I can explain it. I was going to grab you the inxi -Fxxxrz info while things were still broken, but fixed them by accident by installing the latest 5.11 kernel. (I suppose I could still do so if I revert back to 5.8 kernel to see if things break again. Let me know and I'll try it. I assume you're a developer?)

Anyway, just wanted to sum it up to say thanks for the help!

You're welcome. I'm a fellow Mint user just like you. I just have spent quite a bit of time helping others and I learn a lot by doing that. It is possible that if you revert back to the 5.8 kernel--which means the modules will be rebuilt--it might work.

If you consider your issue resolved, please go to your first post in this thread, click the pencil icon, and add [SOLVED] to the title so others know you are no longer seeking help on this issue.

[klrcommute]

I would say there's something more going on with the kernel switching than simply rebuilding the modules ( which I don't know anything about and can only take your word on). The reason I say this is I tried upgrading a couple newer kernel releases in the 5.8 series back when I was trying to troubleshoot this a few weeks ago. Dunno how many iterations of kernel installation/reverting and restarting I went thru, but none of them ever fixed the issue.

I just ran a little experiment to see what happened. With latest nvidia driver version, all updates installed from update manager, etc. I booted into the latest 20.2 Uma build but with the last 5.8.0-63 kernel. Sure enough, low resolution, no hardware acceleration issue remains. Then back into everything with the latest 5.11.0-25 and everything's fixed again. So, I suspect something is going on there somewhere in the last couple months of 5.8 series kernel releases that got resolved in the latest 5.11 series?

[SMG]

I just ran a little experiment to see what happened. With latest nvidia driver version, all updates installed from update manager, etc. I booted into the latest 20.2 Uma build but with the last 5.8.0-63 kernel. Sure enough, low resolution, no hardware acceleration issue remains. Then back into everything with the latest 5.11.0-25 and everything's fixed again. So, I suspect something is going on there somewhere in the last couple months of 5.8 series kernel releases that got resolved in the latest 5.11 series?

I think you are forgetting that the Nvidia driver is proprietary and the people who maintain the kernel you use have nothing to do with Nvidia. There is no way for them to know what is going on with the Nvidia driver and therefore they do not make changes which might affect how the driver performs.

Canonical (Ubuntu) repackages the Nvidia driver so the installation package works properly with Driver Manager and the rest of the code, but they do not have access to the Nvidia driver code. That Nvidia code is not open-source.

It is possible Nvidia has had enough complaints that they have worked with the newer mainline kernels to make sure their driver works better with them than they have in the past and that is why you see a better result with the newer kernel.