00:02mhenning[d]: mangodev[d]: https://archlinux.org/news/linux-firmware-2025061312fe085f-5-upgrade-requires-manual-intervention/
00:03mangodev[d]: mhenning[d]: thank you
00:03mangodev[d]: couldn't find it in a google search, didn't know there was a news post about it ~~because i didn't know arch made news posts like that~~
00:05mhenning[d]: can be useful to subscribe to the arch-announce mailing list https://lists.archlinux.org/mailman3/lists/arch-announce.lists.archlinux.org/
09:47asdqueerfromeu[d]: At least R570 GSP size overhead was offset by further `linux-firmware` splitting :nouveau:
10:32justinrattler: Much of this code we have today we had in another form before. Now kubernetes and such docker code, is basing on chroot namespaces/cgroups and unionfs, but go packages and rust packages arguably like others same as node/electron nuget/mono are bit nicer to deal with, it's just the code written normally isn't as good as old code so to speak innovation or usability isn't present in end user
10:32justinrattler: applications. There is one thing I never understood, certificates can be added authority in it and in firefox and chrome both authority and certs can be added same as system wide, so it's arguable thing that who has authority to ban others, cheat and scam and get away with it. So my real code is uploaded in upcoming years to break such systems and people behind it. I got more vpns added
10:32justinrattler: and undetected however the real innovation is my own code that is thin and easy. The base of networking isn't so bad but it needs less traffic back to back and back and forth for proper systems, so i am looking at utf8, asci, unicode and such along with new generation filesystem specs to develop a better plumbing. this matrix discoord and such stuff is all conspiracy, it offers nothing
10:32justinrattler: but this idiot set banning me out of their rooms. It's value in programming world along with gamja , goguma clients and anything you had ever done is little. all basing on specs that won't function for me aka retarded altogether, retarded people have retarded systems.
13:40mohamexiety[d]: mhenning[d]: gfxstrand[d] what blocks uniform datapath usage in non-uniform control flow? is it a HW issue or NAK not doing it yet?
13:45gfxstrand[d]: The hardware can run multiple paths at a time. So unless we're really sure we know where every lane is, writing to uniform registers is risky because it might stomp other lanes
13:58karolherbst[d]: probably easier to predicate the branches instead
14:00karolherbst[d]: that reminds me, it's something I wanted to look into for the MMA shaders as well, because it should help a little
14:15mohamexiety[d]: gfxstrand[d]: hm I see, thanks :thonk:
14:36gfxstrand[d]: Yeah, more prediction would probably help.
15:56curious-person: hi there, does anyone know if there's a way to reduce GPU power consumption on Kepler era gpus (Quadro K200M) without optimus? Mine is using 12W at idle and I just cannot afford that.
15:57curious-person: K2000m* forgot a zero
15:58curious-person: The reason why I want to avoid optimus is that from what I can tell it's quite a manual process and you basically need to tell optimus on a per-application basis if you actually want to use the GPU, which I'd like to avoid. I'm not even sure if that would come with many savings in terms of power consumption
16:01karolherbst: curious-person: nope
16:01karolherbst: use optimus
16:03f_: How do you measure that?
16:03karolherbst: You might be able to lower the power consumption by 1-2W booting with "nouveau.NvPmEnableGating=1" but there are outstanding issues with that
16:04karolherbst: f_: some GPUs come with power meters and we can read them out
16:04curious-person: f_: I
16:04karolherbst: though we don't properly calibrate them, so the result is a bit... well.. unreliable
16:04f_: fermi apparently doesn't then
16:04f_: Or Powertop is weird
16:04karolherbst: the power meters don't exist on all GPUS
16:04f_: er, *my fermi
16:05curious-person: f_: oops hit enter - I'm using /sys/class/power_supply/BAT0/whatever_now to tell me the watts it's using
16:05karolherbst: anyway.. the optimus stuff exists because it allows for the GPU to be entirely powered off
16:05karolherbst: curious-person: does something show up inside sensors?
16:05f_: ah wait `sensors` does work
16:05f_: https://ircb.dersco.re/uploads/funderscore/5d1c5ba0-f_-paste.txt
16:06karolherbst: yeah, no power consumption tho
16:06f_: yeah
16:06f_: # cat /sys/class/power_supply/BAT0/current_now
16:06f_: 0
16:07f_: well great I'm using zero power, \o/
16:07karolherbst: need to unplug your power supply
16:07f_: yep
16:07curious-person: karolherbst: well it shows me my battery power draw there too lol, but nothing about the GPU
16:07karolherbst: anyway.. the BAT entry in sensors should just do the math for you
16:07karolherbst: curious-person: ah yeah.. guess you are out of luck then
16:07karolherbst: anyway
16:07curious-person: well that's unfortunate, thanks yall!
16:08karolherbst: there is no way to significantly reduce the idle power consumption except with optimus
16:08f_: and optimus is for switching between the GPUs isn't it?
16:08curious-person: I might give nouveau.NvPmEnableGating=1 a try as a last resort, but I guess optimus it is then
16:08karolherbst: yeah.. it should help a bit, but you might run into random bugs...
16:08karolherbst: hard to debug ones even
16:09curious-person: f_: I was trying to avoid the rather manual process of having to enable optimus whenever I need a gpu for something. For one I don't necessarily know which programs do use my gpu and on top of that it's, well, manual work
16:09f_: curious-person: At least you have the option to do it
16:09karolherbst: desktops normally have a context menu entry for it
16:10f_: because I don't :p
16:10f_: (it's either the Fermi or nothing on this laptop)
16:10curious-person: karolherbst: I'm already seeing my first issues just using the dgpu haha, hyprland is drawing a bit weirdly. I'm on wayland too, so this will be fun
16:10karolherbst: oh no
16:11f_: hyprland is very laggy for me when I'm on 2 displays, which means most of the time
16:11curious-person: karolherbst: oh one question too, why config=NvPmEnableGating=1 and not 3? I just found a phoronix post and that mentions that the range goes from 0 to 3
16:12karolherbst: uhm...
16:12f_: karolherbst: Anyway the other day I wanted to see if I could speedrun crashing nouveau, I think I found a way to reproduce my "everything slowly dies" issue
16:12karolherbst: curious-person: shouldn't make any difference
16:12f_: the issue where randomly the 2nd display would freeze or die and the first screen is intact but lagging and switching tty's makes it dead
16:13curious-person: f_: hm not for me, I'm fully on AMD on my desktop and I'm not running into any issues :shrug:
16:13f_: wlr-randr --output DP-2 --custom-mode 3440x1440@165Hz
16:13karolherbst: curious-person: like the 0-3 only existed while development
16:13karolherbst: now it's just a on/off flip
16:13curious-person: karolherbst: phoronix said that the higher you go the more power saving you get, but oh well.
16:13curious-person: the post is from 2018 however
16:14karolherbst: yeah it was the case while this feature was in development
16:20curious-person: huh, gpu power consumption indeed dropped by two watts. Not great still, but a great step in the right direction. I guess I'll have to try optimus then.
16:26curious-person: lol nvm, just noticed that it went right up. seems like the battery was still reporting on the boot process. Seems like no change at all
16:46curious-person: optimus works surprisingly well afterall, thanks again!