[SOLVED] 5.27.7 broke System Monitor GPU sensors plasmoids with NV525

[ShadYoung]

Hi gang.

The problems with NV525 and KDE sensors is well documented. The issue was apparently white spaces. It was eventually fixed in 5.27.1, but seems to be broken again in 5.27.7.

I run 2 dedicated NV cards with quad displays. Card one is being misreported, and card two is showing only "0"s. I was trying to keep the older Nvidia driver (515) because with Plasma 5.26 the Nv525 broke the widgets. However, just prior to the update to 5.27.6 Neon forced an update to 525, which actually worked fine.

Then last week with the 5.27.7 update, they broke again. Attempting to revert back to nv515 broke the system badly (network and input devices stopped working making it impossible to fix on the desktop) and I was forced to timeshift it back a couple of days.

$nvidia-smi reports all the correct values.
NVCP reports the correct values
Psensors reports the correct values
LM-sensors reports the correct values

The desktop monitoring widget does not.

Looking over the threads here, it seems nobody has expressed problems, so it may be something about my setup specifically. Does anybody know how to fix this? Or, failing that, can anybody tell me how to safely revert back to an older Nvidia driver (like 515), as the way I used to do it seems to fail badly.

Because my machine and workflow is so GPU dependent, the loss of this monitoring ability is leaving me in to cold.

[ShadYoung]

Turns out there was a wee minor coding error in ksystemstats gpu plugin module. A simple change from a "> 0" to a ">= 0" in NvidiaSmiProcess.cpp line 199 of the source code fixed the error. Seems it was starting its indexing at one, not zero, which resulted in GPU2's telemetry being output to GPU1, and GPU1s telemetry going off to cyberspace. Adding the "=" to the function started the indexing at zero as it should be.

The fix has been merged and should hopefully be available via updates soon.