00:58 airlied[d]: openrm does appear to boot on it
01:16 airlied[d]: I think the linux-firmware loaders are wrong
01:22 airlied[d]: or maybe I just messed up lcoally, gotta dig some more
01:35 airlied[d]: okay seems that was PEBKAC 🙂
11:43 Anon123: Hi I had a question, in this wikipage https://wiki.gentoo.org/wiki/Hybrid_graphics#Completely_powering_down_the_Nvidia_dGPU , if I follow the instructions for nouveau (blacklisting the nouveau driver, adding the udev rules), my nvidia gpu does disappear from lspci, but it's LEDs still remain on. Do the steps actually power down the gpu then?
11:45 karolherbst: Anon123: why the remove?
11:45 karolherbst: that feels very wrong
11:45 karolherbst: but the bus the GPU is on, also needs to gets `power/control` set to 1
11:45 karolherbst: ehh
11:45 karolherbst: auto I mean
11:45 karolherbst: but it's easier to just load nouveau and let it handle this stuff
11:46 Anon123: There's a way to completely shut it down using nouveau?
11:46 karolherbst: yeah
11:46 karolherbst: just let nouveau load
11:47 karolherbst: there might be other reasons it doesn't, because... distributions have often a poor solution to power mangement
11:47 karolherbst: like if it's a GPU with many subdevices, each driver needs to put the device into "unused" state so it gets powered down
11:47 karolherbst: or well
11:47 karolherbst: not actively used
11:48 karolherbst: nouveau will power up the GPU if youj e.g. attach an external display, or start an application on it
11:49 karolherbst: but of course you can always run into kernel bugs where it's not getting powered down, but then the method explained there won't help you either
11:50 Anon123: right now it seems I couldn't start an application on it even if I wanted to, running __GLX_VENDOR_LIBRARY_NAME=nvidia glxinfo | grep "OpenGL renderer" still returns my igpu : )
11:50 karolherbst: that only works with the nvidia driver
11:50 karolherbst: for nouveau you want to use "DRI_PRIME=1 glxinfo"
11:51 karolherbst: some desktops allow you to launch app on a specific device via a context menu, but the situation with where to launch an application on ain't the greatest atm
11:52 karolherbst: might also want to check dmesg if anything went wrong
11:53 Anon123: Oh yeah sorry, running it with DRI_PRIME=1 returns a string of only NV176, is that all that it's supposed to show? I was expecting something more verbose?
11:53 karolherbst: it will be when using zink
11:54 karolherbst: ehh wait..
11:55 karolherbst: Anon123: try "NOUVEAU_USE_ZINK=1 DRI_PRIME=1 glxinfo"
11:55 karolherbst: should get you better performance with games and stuff anyway
11:55 karolherbst: though might want to only use it if you are on mesa 24.2 or newer
11:56 magic_rb[d]: Since you have two gpus, id set `NOUVEAU_USE_ZINK` globally, and personally if i could id run linux latest and mesa master (im blocked by zfs, you probably arent)
12:02 Anon123: Tried using those parameters, but there'sno difference it glxinfo's output, is that normal?
12:02 karolherbst: uhh.. depends on your mesa build
12:03 karolherbst: could either be too old or zink not being enabled in the driver
12:03 karolherbst: in any case, "NV176" is the chipset and should be ga106
12:04 karolherbst: which is ampere
12:04 karolherbst: which means your setup should be in order and the GPU _should_ be put ti sleep if nothing else goes wrong
12:05 Anon123: If only there was a way to check....
12:06 Anon123: Anyway thanks for the help
12:06 karolherbst: what do you mean "it's LED's still remain on" btw
12:06 karolherbst: is this a desktop or laptop?
12:07 karolherbst: or do you have an LED indicator?
12:07 karolherbst: anyway, you can check with `grep . /sys/bus/pci/devices/*/power/runtime_status` which PCI devices are powered on/off
12:07 karolherbst: or supposed to be powered on/off
12:08 karolherbst: the firmware is involved with actually powering the devices off, so it kinda depens
12:08 karolherbst: the GPU and its subdevices should be "suspended" and also the PCIe host bridge it's connected to
12:09 Anon123: this is a dekstop, it's LEDs just mean the fluff that comes with 'gaming' cards these days
12:09 karolherbst: ahhh...
12:09 karolherbst: well
12:09 karolherbst: if it's a desktop you are probably out of luck, because I'm not aware of any desktop firmare actually cutting power to the GPU
12:09 karolherbst: in which case
12:10 karolherbst: having a driver loaded reduces power draw, because then the GPU can actually get put into a lower perf state instead of the default one, which are usually not the lowest power state
12:10 karolherbst: the nvidia driver or nouveau with GSP enabled should do the trick
12:11 karolherbst: Anon123: you might need to boot with `nouveau.config=NvGspRm=1`
12:11 karolherbst: which will enable power management + gives you more perf when you need it
12:11 Anon123: I've done that already
12:11 karolherbst: okay, then everything should be set
12:12 karolherbst: but yeah.. maybe the wiki should get updated mentioning that it's pointless to follow on a desktop :D
12:12 karolherbst: the one you linked above
12:12 Anon123: (don't know how to check if i did it correctly, cat /proc/cmdline has it in its output though)
12:12 Anon123: Yeah I'll see if I can get the wikis edited
12:13 karolherbst: you could try out a game and if the performance is good enough it's probably enabled
12:14 karolherbst: though I think "journalctl --dmesg | grep -i nouveau" might print something but I'm not 100% sure
12:14 karolherbst: might also depend on the debug level set
12:14 Anon123: Ok yeah according to lspci my nvidia card's id is 01:00.0 and apparently I get the line `/sys/bus/pci/devices/0000:01:00.0/power/runtime_status:active`, that's the nvidia card right?
12:14 karolherbst: yeah
12:14 karolherbst: probably
12:14 karolherbst: it depends
12:14 karolherbst: lspci should tell
12:14 karolherbst: but that stuff won't be enabled on a desktop anyway
12:15 karolherbst: or at least not by default
12:15 karolherbst: maybe you want to use a udev rule like this: SUBSYSTEM=="pci", TEST=="power/control", ATTR{power/control}="auto"
12:16 karolherbst: but could be that some drivers can't handle it
12:19 Anon123: Separate question, but is this a cause for concern: nouveau 0000:01:00.0: [drm] Cannot find any crtc or sizes (from journalctl --dmesg). Mind you I don't actually intend to use the card for anything
12:19 karolherbst: nah, you can ignore that line
12:20 karolherbst: Anon123: if you don't use it for anything, you mean on linux and you have a dual boot setup? Because otherwise it might be best to just remove the card
12:21 Anon123: Yes I have a dual boot setup
12:24 karolherbst: okay.. yeah, I'm not quite sure how you could have the benefit of power management to put the GPU into a low power state and also prevent apps from using it. If you only care about apps not using it, you can certainly disable the driver from loading. If you also care about the GPU drawing as few power as possible, then you'll need to have a driver loaded
16:37 marex: Hi , I ran into https://gitlab.freedesktop.org/drm/nouveau/-/issues/381 also on 880M today, with debian 6.10.4
16:38 marex: I can trigger this on sway when using pavucontrol
16:38 marex: I dont think it was there in 6.9.y , but I might be wrong
16:38 marex: now looking at the git log between 6.9 and 6.10 and so far I don't see anything obvious
16:39 marex: and there is a visual corruption when I use drop-down menu in pavucontrol
16:42 marex: hum, it is in 6.9.y too
20:02 karolherbst: yeah.. sounds like you just get unlucky somtimes
23:41 marex: karolherbst: I did find a workaround it seems, but ugh
23:41 marex: karolherbst: that 880M is a gift that keeps on giving, really