#asahi-dev · 2023-04-22 04:17
chadmedother SIP blocks sucking down more juice than they should#asahi-dev · 2023-04-22 04:17
chadmedthe divergence between system draw and ap draw at idle tells me that theres other things we need to do to bring down idle power use#asahi-dev · 2023-04-22 04:16
chadmedi think realistically we're at the limit of what we can achieve by clock gating #asahi-dev · 2023-04-22 04:16
chadmedthere are probably superfluous wakeups being caused by other stuff, because i managed to ssh into a partially failed boot before and idle was 2.2 W #asahi-dev · 2023-04-22 04:15
chadmedidle power doesnt really change if i kill plasma and go to a tty#asahi-dev · 2023-04-22 04:10
chadmedTEO parks them in the lowest pstate but it also isnt great for interactivity so theres our tradeoff#asahi-dev · 2023-04-22 04:09
chadmedthe menu governor parks firestorm in the 6th pstate to idle them, which is going to cause idle power to be higher through leakage or watever#asahi-dev · 2023-04-22 04:08
chadmedi think the next "free" bit of desktop use power draw is definitely going to be doing something about the touchpad interrupts on non-mtp machines#asahi-dev · 2023-04-22 04:07
chadmednot too bad but theres probably more we can do#asahi-dev · 2023-04-22 04:07
chadmedok we're down to 2.55 W total AP idle and about 5 W system with the screen at the minimum brightness#asahi-dev · 2023-04-22 02:28
chadmedat least now that i know we're still getting perfectly polynomial power scaling on the cores outside of turbo mode i can do some stuff to compensate for the noise in the readings at low power#asahi-dev · 2023-04-22 02:26
chadmednot sure coremark is really hammering the core as much as it could #asahi-dev · 2023-04-22 02:26
chadmedstill, 3 W dynamic for a firestorm core at full flog seems very low #asahi-dev · 2023-04-22 02:26
chadmedif i take the turbo states out of the dataset then the numbers line up perfectly with what youd expect#asahi-dev · 2023-04-22 02:25
chadmedless dynamic power coming from idle cores#asahi-dev · 2023-04-22 02:25
chadmedoh yeah it does ok thats why the numbers are different#asahi-dev · 2023-04-22 02:24
chadmedWFI halts the clocks on CPUs where it is issued until theyre woken by an interrupt right#asahi-dev · 2023-04-22 00:00
chadmedso again, something that we can probably beat macos at#asahi-dev · 2023-04-21 23:59
chadmedone funny thing about macos's implementation is that it ends up saturated at max volume and spills out#asahi-dev · 2023-04-21 23:57
chadmedbut theres still a risk of overexcursion that would just go away completely if we could cut out the "real" bass#asahi-dev · 2023-04-21 23:57
chadmedi think ive changed my mind on its absolute necessity though, ive done some tricks to get the bass to sound good without it #asahi-dev · 2023-04-21 23:56
chadmedi think you should be able to do this with a funky chain of biquads if you want to have a go at it#asahi-dev · 2023-04-21 23:55
chadmedas the yaks page explains, basically what we do is split off some range of bass frequencies with a bandpass filter, amplify the harmonics, then mix those back into our signal #asahi-dev · 2023-04-21 23:54
chadmedKnedlik: psychoacoustic bass extension is kind of esoteric so no biggie that youre unfamiliar #asahi-dev · 2023-04-18 12:18
chadmedbut higher#asahi-dev · 2023-04-18 12:18
chadmedresolution doesnt affect it so its still cpu bound #asahi-dev · 2023-04-18 12:16
chadmedmight be the sus eas numbers, but im getting ~3700 by default whereas last time i did the testing i barely got 2300 with the performance governor and pinning glmark2 to the pcores#asahi-dev · 2023-04-18 12:01
chadmedsame mesa #asahi-dev · 2023-04-18 12:01
chadmedoh in other news glmark2 scores have doubled for no apparent reason???#asahi-dev · 2023-04-18 11:45
chadmedill do the mini tomorrow now, its getting late#asahi-dev · 2023-04-18 11:45
chadmedj314 atm#asahi-dev · 2023-04-18 11:20
chadmedkde reporting ~7h worth of battery watching 1080p30 vp9#asahi-dev · 2023-04-18 11:19
chadmedbut thats what the SMC tells me so...#asahi-dev · 2023-04-18 11:18
chadmed3.3 W per core at the top firestorm pstate seems... low#asahi-dev · 2023-04-18 11:18
chadmedhm my numbers have changed since last time and i think theyre wrong...#asahi-dev · 2023-04-18 10:05
chadmedlets see just how bad they are heh#asahi-dev · 2023-04-18 10:01
chadmedoh wow idle power is closer to 8 W now, im assuming EAS doesnt like pstates without opp-microwatt#asahi-dev · 2023-04-18 09:59
chadmedlet me cobble together enough hwmon to do opp-microwatt measurements#asahi-dev · 2023-04-18 09:58
chadmedawesome, boost pstates are working #asahi-dev · 2023-04-18 09:37
chadmedi assumed that the device in pro audio mode is exposed as a single channel by pipewire#asahi-dev · 2023-04-18 09:36
chadmedhttps://github.com/chadmed/asahi-audio/tree/desktops/mini i got a bit distracted oops#asahi-dev · 2023-04-18 08:22
chadmedalmost 12 hours off charge and only 60% of the battery gone#asahi-dev · 2023-04-17 21:17
chadmedif the codec s2idle fixes get merged ill start testing opportunistic sleep etc and see how that changes things #asahi-dev · 2023-04-17 21:16
chadmedplasma reading 10 hours of battery as im typing this and wiggling the cursor where it used to be ~6 but really closer to 4#asahi-dev · 2023-04-17 21:15
chadmedalso t6000 ~11 hours of idle time with the screen off where it used to be less than 7 so theres a real improvement#asahi-dev · 2023-04-17 21:14
chadmedwe dont do anything with MCC pstates yet do we? can it be switched down to its lower pstate when all cores are in low pstates or something like that? #asahi-dev · 2023-04-17 12:41
chadmedits gotta be more than that surely, at least on this machine the thing is stone cold now idling and watching 1080p30 video barely warms it#asahi-dev · 2023-04-17 12:39
chadmedi had some patches that added enough hwmon to get the key that measures that#asahi-dev · 2023-04-17 12:39
chadmedi want to look at how much power the cores are pulling down without the noise of the rest of the system #asahi-dev · 2023-04-17 12:35
chadmedive lost my smc power meter patches so ill rebuild those and do some more testing later