Set NVreg_DynamicPowerManagement on affected models

[crawfxrd]

Move this setting out of system76-power.

Add addw4 to the list of models that needs it to avoid issue where GSP firmware crashes on GPU power-on after GC6.

[crawfxrd]

Something (probably the driver) writes the option in /usr/lib/modprobe.d/nvidia-runtimepm.conf. This probably has to be removed as well? Or does /etc/modprobe.d have any guarantee to override /usr/lib/modprobe.d?

It's fucking ubuntu-drivers-common again.

[leviport]

Does this need a ./bump-version.py ?

[crawfxrd]

Yeah. I'll fix it when I figure out the modprobe rules and if I need to patch u-d-c again.

[robotman40]

As I said in the other issue, I tried this and it didn't help.

I'm on a CachyOS install though.

[jacobktm]

the DynammicPowerManagement nvidia driver parameter still ends up set to 2 after installing this along with the PR in system76-power: pop-os/system76-power#465

It looks like the setting NVreg_DynamicPowerManagement is still being set to 2 in /etc/modprobe.d/system76-power.conf and that value is taking precedence.

[crawfxrd]

system76-power graphics <mode> would need to be run so the /etc/modprobe.d/ file is updated.

[jacobktm]

This does resolve the GSP crash in the addw4 and I also did some testing in an oryp12, bonw16, and gaze20 and everything works normally in those systems.

[robotman40]

I ran this on my CachyOS Addw4 and it seems to have fixed the issue with the GPU falling off the bus, as well as the suspend problem.

But I still think the suspend problem could be solved at a firmware level as it doesn't work on Windows.

[robotman40]

I ran this on my CachyOS Addw4 and it seems to have fixed the issue with the GPU falling off the bus, as well as the suspend problem.

But I still think the suspend problem could be solved at a firmware level as it doesn't work on Windows.

Update: It seems that the open Nvidia drivers just fall off the bus immediately. In general though, the open drivers were problematic, even before this as like without this update, DX12 games wouldn't run.

[robotman40]

I ran this on my CachyOS Addw4 and it seems to have fixed the issue with the GPU falling off the bus, as well as the suspend problem.
But I still think the suspend problem could be solved at a firmware level as it doesn't work on Windows.

Update: It seems that the open Nvidia drivers just fall off the bus immediately. In general though, the open drivers were problematic, even before this as like without this update, DX12 games wouldn't run.

And it seems that even with the proprietary drivers, the GPU still likes the fall off the bus rather quickly.

So this isn't really fixed...

[robotman40]

After spending a whole day with this on my addw4 running CachyOS, I discovered this:

The issue was basically not solved.

I am not QA or anything, but @jacobktm, could you verify if this update really fixed the issue? I know you guys test on Pop!_OS and Ubuntu, but I feel like patching this on an OS level isn't really addressing the issue. Like the suspend issue still happens on Windows and on most Linux distros, the GPU will fall off the bus, even with this new update.

We really want this issue to be solved as the lack of usable Nvidia graphics is a major deal breaker to me.

[leviport]

This wasn't the full fix. We are still working on it.

[robotman40]

This wasn't the full fix. We are still working on it.

Oh I see, my apologies for misunderstanding

[robotman40]

This wasn't the full fix. We are still working on it.

Is there any update or ETA on when the GPU bug will be fixed? It's been about a year and a half and no fix has come around for this.

[leviport]

Have you tried the latest firmware update?

[robotman40]

Have you tried the latest firmware update?

I just installed it and it seems to have fixed the problem. I did have to manually set NVreg_DynamicPowerManagement to 0x01 at /usr/lib/modprobe.d/nvidia.conf on my CachyOS install however. Is there a way to get /etc/modprobe.d/nvidia-runtimepm.conf to override the former?

Also, using the open drivers on any distros still makes DX12 inoperable. Any fix for that?

Besides, the sleep issue still persists on Windows, meaning a firmware-level fix is still needed. It only is fixed on CachyOS for me.