7900XTX core waken when monitoring

[Numerlor]

I'm trying out having my monitors plugged into my (amd) igpu's ports and doing passthrough from dgpu. At first after switching the cable the TBP of the dGPU was constantly at <5W but after restarting a couple times and messing with some GPU settings, which were then rolled back and the driver reset to stock, the GPU core is now periodically woken by hwinfo; all wake disabled gpus/poll sleeping gpus options are turned off.

I'm not sure if this is necessarily an issue with how hwinfo takes the values as it seemingly worked before but I'm lost as to what's causing it to suddenly be woken when it didn't do that before.
The image from hwinfo shows the power spikes that weren't there before, the second image is from Process Hacker that shows up gpu usage spikes that start only when hwinfo or adrenaling logging is turned on, and depend on the interval of either of them

 

[Martin]

HWiNFO checks in each cycle whether the dGPU is sleeping and if yes, it doesn't poll it. So I think that when you changed settings and reset the driver something went wrong there and it's no longer properly indicating the sleeping state.

 

[Numerlor]

I suppose I should go on a hunt to figure out what I had enabled in the previous driver install then as my second barebones OS also has this behaviour along with the new clean driver I tested it with. But it doesn't really make sense to me as the last changes I did were just some registry things I deleted afterwards and I'm not even sure if the driver reads them for this gen of gpus.


Does hwinfo completely skip polling if the GPU is sleeping? Before I started seeing the spikes I still had sensor values on the GPU, but the core was just constantly at 0W and 0 MHz clocks, and a couple W on memory and SOC, while it now gets clocks bumped up. I can't effectively tell whether the clocks are completely idle without hwinfo but at least my PSU's sounds seems to indicate it as it's more whiny at lower power and the spikes on the images correspond to when it stops whining for a bit

 

[Martin]

Yes, HWiNFO should completely skip polling in that cycle.

 

[Numerlor]

Thanks that's good to know, so it was being actively monitored before too but the core was unaffected unlike now, or it happened in a way that the monitoring skipped it. I'll try asking AMD about it though I'm not expecting much there from normal support