NVIDIA dGPU with fine-grained RTD3 support could not be turned off due to occupation from plasmashell

Greetings everyone! As a long-time KDE user, it has been a pleasant ride these 7 or 8 years. However, I’ve found something a little bit annoying to me recently.

Here’s my laptop’s specs:

  • i7-10750H with Intel graphics, modesetting driver
  • NVIDIA Quadro T2000 Max-Q (basically GTX 1650Ti Max-Q, which supports RTD3 power management) with closed-source drivers version 550.144.03 installed according to the guide from openSUSE wiki
  • openSUSE Tumbleweed, newest snapshot possible (kernel 6.13 series)

After the upgrade from Plasma 6.2 to 6.3, the NVIDIA GPU could not be turned off even with fine-grained RTD3. This still worked in Plasma 6.2, which I checked by reverting to the snapshot with the particular version (kernel 6.12.8 in that snapshot). The power status is checked from:

cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_status

where it is always active (on battery or AC doesn’t matter, and could turn into suspended in 6.2), and:

 cat /sys/bus/pci/devices/0000:01:00.0/power/runtime_suspended_time

where the suspended time doesn’t accumulate.

Referring to Archwiki and Arch Linux forum where an environment variable of:

__EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/50_mesa.json
__GLX_VENDOR_LIBRARY_NAME=mesa

and X11 configs to run XWayland on solely iGPU doesn’t fix the problem.

By occasion I’ve checked

 fuser -v /dev/nvidia*

and gives the output:

/dev/nvidia0:        x    3677 F...m plasmashell
/dev/nvidiactl:      x    3677 F.... plasmashell
/dev/nvidia-uvm:     x    3677 F.... plasmashell

It seems that in Plasma 6.3 (and 6.3.1), plasmashell is occupying the dGPU even if it’s not in use in anyways, which kills my battery really fast (since it throws an even larger burden to it alongside the power-hungry Intel CPU).

I’ve not sure if it’s a bug, nor do I know how to debug (lacking correspondent knowledge to this). Any ideas on the issue and possible guide to debug it are welcome from my perspective. Thanks in advance!

Have you tried disabling GSP firmware?

The use of the GSP firmware, enabled by default since version 555 of the NVIDIA driver released in June 2024, is known to cause a range of issues including Vulkan failures and system crashes.

No (just tried), and I don’t think that’s the issue. AFAIK the GSP bug doesn’t affect driver 550.