00:24 Venemo: Roberth1990: does that only happen with kde plasma or also other desktops?
00:26 Roberth1990: Venemo: doesn't happen with nvidia and plasma, but yeah I need to test with other desktops
00:29 Roberth1990: brb
00:59 Roberth1990: Venemo: seems to be there with gnome too, but the worst cases is sometimes impossible to reproduce
01:06 Roberth1990: yeah its there with gnome too definitively
01:11 Roberth1990: tried to look at amdgpu bugs for a similar issue, but the issues I was able to find seemed to include the mouse cursor too, this issue only affects mpv, firefox with hardware acceleration: for example scrolling facebook and games
01:11 soreau: Roberth1990: It sounds like you may very well have found a driver bug. Maybe the most useful path forward is to file a mesa issue
01:11 Roberth1990: sure its a mesa issue? it happens with both radeonsi and radv?
01:12 soreau: well it might not be a mesa issue per se, but the mesa issue list gets a lot of exposure by many developers and they should be able to redirect you further if need be
01:12 soreau: it could be a drm/kernel problem
01:13 Roberth1990: hmmm
01:13 soreau: btw, have you checked kernel logs for anything interesting/suspicious?
01:14 Venemo: Roberth1990: do you have anything suspicious in dmesg or lspci?
01:15 Roberth1990: [ 5.116131] amdgpu: Overdrive is enabled, please disable it before reporting any bugs unrelated to overdrive.
01:16 Roberth1990: nothing in dmesg
01:17 soreau: what's the output of `cat /proc/cmdline`?
01:17 Roberth1990: https://paste.cachyos.org/p/8e14588.txt
01:17 fililip: Roberth1990: what resolution is your display?
01:17 Roberth1990: fililip: 3840 x 2160
01:18 fililip: ah, could it be this? gitlab.freedesktop.org/drm/amd/-/issues/4753
01:18 fililip: see if it's tied to memory clock changing
01:18 Roberth1990: fililip: https://paste.cachyos.org/p/d398a8c.txt
01:19 fililip: oh, sorry, I pasted the URL to the issue wrong, here it is again: https://gitlab.freedesktop.org/drm/amd/-/issues/4753
01:19 Roberth1990: none of the kernel parameters have affected the issue FYI
01:19 fililip: I experience stutter across the whole desktop session for a brief moment when the memory/vram clock changes from one state to another
01:19 Roberth1990: fililip: maybe
01:20 fililip: and I use 3840x2160@160 + 1920x1080@120 screens with dp1.4
01:20 Roberth1990: 3840x2160@120 here
01:20 Roberth1990: changing it to 60 hz made even the cursor stutter
01:21 fililip: it should not stutter at all with one display at 60Hz though if it's the same issue as mine from my testing, but you could verify this by trying to manually set the vram clock state
01:22 fililip: is it possible for you to try the steps from this comment: https://gitlab.freedesktop.org/drm/amd/-/issues/4753#note_3228433 to see if it does the same thing?
01:22 Roberth1990: I seem to reproduce it by first play with mpv, not much stuttering, close mpv, scrolled a bit with firefox, then went started mpv, and bahm tons of stutter
01:24 Roberth1990: that website doesn't respond, brb
01:38 Roberth1990: if I understand it correctly, that vsync test fails regardless
01:39 fililip: even when you set manual DPM and force a certain memory clock state?
01:42 Roberth1990: yeah, if I switch tab in firefox and go back to the test, the vsync logo seems fine for a new second before flashing like crazy
01:43 Roberth1990: if I understood it correctly on how the test works
01:43 fililip: fwiw, it doesn't have to be that website + chromium specifically, you can just run anything that's constantly updating in the background and observe it when changing memory states
01:44 soreau: isn't there a way to force a certain state so it doesn't change, for testing to verify this is the issue?
01:44 Roberth1990: well when I change states it doesn't make mpv mistime/delay frames
01:46 Roberth1990: but im not sure if im doing to the right card
01:46 Roberth1990: https://paste.cachyos.org/p/a279064.txt
01:47 fililip: that's at 120Hz?
01:47 fililip: there's just one card here so it seems fine
01:49 Roberth1990: yeah
01:49 Roberth1990: but wait
01:50 Roberth1990: every time I do it, change state, a frame is delayed in mpv
01:50 fililip: alright, that sounds like the issue
01:51 Roberth1990: but I cant provoke any mistimed frames
01:51 fililip: and once you set a state, say 456MHz, and leave it, does it stutter at all anywhere?
01:51 Roberth1990: uhhh
01:51 Roberth1990: how do I see what frequency each state represents?
01:52 fililip: you can do cat /sys/class/drm/card1/device/pp_dpm_mclk
01:52 fililip: and amdgpu will print all possible states
01:55 Roberth1990: seems fine at state 1 yes, refresh rate seems to be much stable too
01:55 Roberth1990: what mpv estimates
01:55 fililip: it should be stable at any state, try setting 4 or 5 even
01:56 Roberth1990: hmm some delayed frame came, but it was related to pausing video/alt tabbing in and out of the window it seems
01:57 fililip: if you're on KDE then alt+tab with the default window switcher will cause desktop-wide stutter unfortunately, it's a Qt bug
01:58 Roberth1990: seems stable at 5 too
01:59 Roberth1990: fililip: yeah I figured that must be a completely different issue, but it's the same symptom pretty much
01:59 Roberth1990: now when I alt tabbed back to mpv and resumed playback the stutter became extreme, powerpoint slide
02:00 Roberth1990: over 40 mistimed frames before going back to normal
02:01 Roberth1990: but otherwise stable
02:02 fililip: I found RDNA4 to have an issue where memory clock changes (which happen automatically if you don't touch anything and just boot into Linux without GPU tweaking tools like LACT/CoreCtrl/etc.) cause the vertical blanking interval to be extended to allow reclocking, causing stutter at high resolutions and refresh rates
02:02 fililip: that's one issue, the other one I'm not sure about
02:09 Roberth1990: well guess Ill just use my nvidia gpu until there is an sollution
02:10 fililip: if you'd like a workaround, you can use LACT (since you're using CachyOS you should be able to install it via pacman just fine with pacman -Syu lact) to limit the clock speed to a fixed state, and even create profiles to quickly change between fixed state → automatic reclocking for games
02:11 Roberth1990: id rather just use the nvidia gpu for now
02:17 Roberth1990: but i have commented on the issue
02:22 fililip: thanks!
02:24 Roberth1990: thanks to you too btw I hate to file bugs
03:03 Venemo: Roberth1990: can you please show us your sudo lspci -vvv and sudo lshw
03:04 soreau: kinda makes me wonder why this is a problem seemingly only on this particular gpu set
03:07 Venemo: bad pcie settings, gpu badly seated, wrong pcie slot, or just being an egpu, could cause this sort of issue
08:57 Roberth1990: Venemo: https://paste.cachyos.org/p/64ff092.txt https://paste.cachyos.org/p/8d92f09.txt
09:57 Roberth1990: Venemo: it is in the right pci-e slot, the only problem is that it's an pci-e 4.0 slot
12:17 fililip: Venemo: it's just a display code/firmware issue, if you have VRR set to auto in KDE, you can see your screens dropping their refresh rates when reclocking is needed
12:18 fililip: but interestingly, I'm also on a pcie 4.0 chipset
12:21 fililip: it also doesn't happen at single-display 1440p or 1080p though, at least on my setup
14:23 Venemo: fililip: sorry, what's your issue?
14:34 fililip: it's https://gitlab.freedesktop.org/drm/amd/-/issues/4753, when you're using a 2160p display at more than 60Hz, memory clock speed changes cause stutter events due to display vblank stretching
14:35 fililip: I remember you said you would have an RDNA4 card mid-January, do you have it now to test this?
14:37 Venemo: I will get back home next week
14:38 fililip: this turned out to be the issue that Roberth1990 has
14:43 Venemo: aha
17:33 fililip: Venemo: I found a workaround for the linux 6.18+ gfx12 compute queue timeout bug, RADV_DEBUG=nocompute seems to do the trick
17:34 fililip: does that envvar fix CI for mesa?
17:47 Venemo: fililip: I don't know. better would be to just bisect the kernel and fix it
17:48 fililip: it's a shame it's difficult to trigger in the first place, I could try bisecting if I could figure out what causes it :(
19:08 Roberth1990: great the extreme stuttering happening sometimes in mpv also happens in games too
19:12 fililip: and if you force a memory clock setting like before it doesn't stabilize when you're playing the game?
19:20 Roberth1990: kind of hard to test, it happens after hours of gaming every time
19:21 fililip: ah then it can't be that issue, it has to be something else
19:21 Roberth1990: yeah its so extreme, vram clock doesn't multiple times per second
19:22 fililip: I know that night light can cause stuttering when a compositor gradually adjusts it, but this doesn't sound like it
19:22 Roberth1990: I dont use that
19:23 fililip: does it only happen when you're pressing keys and moving the mouse or is it not related to that?
19:24 Roberth1990: well the is not very graphics intensive, and it happens even when all the AI stuff is paused
19:24 Roberth1990: but yeah I have to move around the screen on the map for it to triggered ti seems
19:25 fililip: the issue I'm thinking of is a Steam issue, try setting LD_PRELOAD="" %command% in launch options (if you're using Steam to play the game that's affected)
19:26 Roberth1990: but same happens in mpv sometimes too
19:26 Roberth1990: *the same
19:27 Roberth1990: its one of two extreme scenarios that happens with both, either this extreme stuttering, its like someone is shaking the monitor, or its a powerpoint slideshow
19:29 fililip: that could potentially be yet another thing, but the issue where moving the mouse or pressing and holding a key causes consistent stuttering is fixed by setting that launch option, try to see if that helps with games
19:29 Roberth1990: I dont run the game through steam unfortunatly
19:29 fililip: oh, ok, so that can't be it then
19:29 Roberth1990: but moving the mouse or holding a key doesn't trigger it
19:31 Roberth1990: and as I said the exact same thing happens in mpv too
19:37 fililip: is it possible for you to share your whole dmesg log since boot? feel free to redact anything you deem sensitive
19:39 Roberth1990: https://paste.cachyos.org/p/4d288a3.txt
19:39 Roberth1990: not sure if that is everything
19:46 fililip: hm, ok, I don't see anything out of the ordinary here, is anything running out of memory on your system?
19:49 Roberth1990: nope
19:50 Roberth1990: it never happend with the nvidia gpu
19:51 fililip: did the nvidia card also have 8GiB of vram?
19:54 Roberth1990: nah 12 gb
19:54 Roberth1990: but the game and mpv is not vram hungry
19:58 Roberth1990: 6.6 gb of vram used when playing the game
22:23 soreau: Roberth1990: it almost sounds like a hardware issue. have you checked operation of the gpu fan(s) and temperature of the gpu? have you tried visually inspecting it and reseating it if those are options?
22:24 soreau: and if mpv is using the gpu for decoding/playback somehow, have you tried mpv options to make it use only cpu?
22:43 Roberth1990: mpv is not using it for decoding, its using it for rendering
22:45 soreau: are you using wayland or X?
22:46 soreau: You mentioned plasma and gnome, so I assume wayland, in which case you could try `mpv -vo wlshm video.ext`
22:50 soreau: but you said it doesn't happen using mpv when using nvidia so I'm guessing that command would only show that it happens only with gpu video output and not with cpu shared memory
22:59 Roberth1990: https://i.imgur.com/VTxowDH.png
22:59 Roberth1990: yeah wayland
23:00 Roberth1990: uhh throttling power?
23:06 Roberth1990: "MangoHud may incorrectly display throttling power even when your GPU is functioning normally."
23:12 Roberth1990: soreau: https://i.imgur.com/0iysNwC.png
23:13 Roberth1990: that was with displayresample though
23:14 soreau: are you using mangohud? universally in the compositor or something?
23:14 Roberth1990: works fine without it, but its sometimes hard to reproduce
23:14 Roberth1990: just used mangohud now to see if there was anything noteworthy to observe
23:15 soreau: the test was supposed to be just mpv with wlshm, not with displayresample and mangohud :P
23:15 Roberth1990: not using mangohud in mpv, that is in a game
23:15 soreau: oh
23:16 soreau: well there is GALLIUM_HUD FWIW
23:16 soreau: Roberth1990: so I didn't really catch the outcome. Does wlshm have the problem too?
23:17 Roberth1990: its hard to say, its not easy to reproduce
23:18 soreau: how about `sensors` to check gpu temp? are the fans on the gpu working?
23:19 Roberth1990: look at the mangohud screenshot
23:19 soreau: I don't really know what I'm looking at/for there
23:20 Roberth1990: that screenshot of mangohud is from while playing a game?
23:23 Roberth1990: tons of stuttering with wlshm too
23:24 Roberth1990: for some reason mangohud shows info the radio mpd plays at the bottom, just to clear up that confusion
23:38 soreau: ok so if wlshm stutters too, maybe the issue is (also) affecting the display server(s), i.e. plasma/gnome
23:40 soreau: just seems like the problem is at a drm/kernel/hw level, so I'd try 1) checking `sensors` for gpu temp 2) if the temps are high, check the fans on the gpu 3) reseat the gpu, and inspect it carefully in the process