13:07 MrCooper: agd5f_: per https://gitlab.freedesktop.org/mesa/mesa/-/issues/14451 , the GPU clock reported by the kernel seems ~0.2% off for at least some APUs; any idea what the reason might be?
22:31 colo: I have a Radeon RX 6600 in this system here, and the associated /sys/class/hwmon/hwmon1/power1_cap says board(?) power is capped at 100W. this seems lower than what the AMD drivers told me about other RX 6600 cards under Windows - is something wrong with this particular card?
22:33 colo: https://paste.debian.net/plain/1414251
22:34 colo: better/more system info: https://paste.debian.net/1414252/
22:54 johnny0: 100W sounds right, the 6600 family varied a bit in their power limits
22:54 johnny0: if the card has dual vbios one might have a higher power limit than the other
23:09 colo: johnny0: under Windows, the power consumption is calculated to be around 130W peak during benchmark runs. I'm not aware of a secondary VBIOS on this low-end card (Sapphire Pulse)
23:13 johnny0: interesting, was the expected limit present in an earlier kernel?
23:14 johnny0: you could dump the powerplay table and examine it with something like 'upp'
23:17 colo: johnny0: I don't know about earlier kernels with that card, but I can check tomorrow
23:17 colo: upp is https://github.com/sibradzic/upp I presume?
23:18 colo: I will play around with that, too. thanks very much for the pointer!
23:18 Venemo: colo: do you actually get lower performance or is it just the power cap number that is different?
23:20 colo: Venemo: I don't know :) I just see an apparently GPU-bound workload running (my wife playing a steam game next to me) and the power readings nvtop produced seemed low to me. I had the same card (same vendor/model, but not the very same physical entity) in another system running Windows maybe a year ago, and remember it using more power there.
23:21 Venemo: colo: I see. well, it's hard to judge without seeing benchmark results
23:25 colo: the dumped table is here, ftr: https://paste.debian.net/hidden/cfe9fc7b/
23:26 colo: I have the other card I talked about around somewhere still, will try to reproduce this experiment with it around christmas
23:27 colo: super interesting tool, thanks again johnny0 :)
23:31 Venemo: colo: your link says, TotalBoardPower: 97
23:31 Venemo: so, 100 W seems appropriate
23:45 colo: yes. looking forward to learn what the other card's data says :)
23:46 colo: thanks for your help :)