06:45 qdel: Hi, New to nouveau, i finally dropped bbswitch to try prime offloading on my GT750M optimus laptop. It work quite nicely (at least i can suspend my laptop \o/). But i have issue with reclocking. card hard freeze most of the time i want to leave slowest pstate. Is there any up-to-date chart about how this feature is enabled?
06:45 qdel: i can only find info accross websites which are a couple of years old.
08:31 mupuf: karolherbst: I guess what you are seeing is the FSRM coeffiecient being tweaked by the BIOS
08:31 mupuf: I guess it will be easy to check
09:03 karolherbst: mupuf: maybe
09:03 karolherbst: mupuf: can it also react to power consumption?
09:05 mupuf: yes!
09:05 mupuf: the computed one
09:05 mupuf: not the one from the sensor
09:20 karolherbst: I see
09:20 karolherbst: makes sense then
09:21 karolherbst: mupuf: so I can assume, that the hardware is able to "read"/"calculate" the computed power consumption
09:21 karolherbst: which kind of gets me the feeling, we could read it out somehow
09:21 mupuf: dude, after all these years, when will you actually read the papers I asked you to read :D
09:24 karolherbst: mupuf: well, I also know since yesterday, that we have a todo wiki page
09:25 karolherbst: but actually I am sure I read your paper I just forgot about most of it
09:27 Tom^: its like ikea furniture, and a born fault. we men simply does not read instructions!
09:27 Tom^: mupuf: dont know if you saw yesterday or was it the day before that, but i have a 780ti sort of working wanting a new owner if you are intersted, il pay the shipping.
09:28 mupuf: Yes, I read that. I am sliglthly concerned about the "sort of working" :D
09:28 mupuf: where would the shipping be from?
09:29 mupuf: and what chipset is that?
09:29 Tom^: from sweden
09:29 Tom^: and uh yeah, it works. it just randomly downclocks when using it for 100%
09:29 Tom^: like it where thermal throttling, but its only around 50C
09:29 Tom^: :D
09:30 Tom^: doesnt happend all the time either
09:30 Tom^: chipset, uh isnt it gk110b if my memory serves me right.
09:30 mupuf: oh, sounds good then
09:30 karolherbst: mupuf: fun part is, it only happens with nvidia
09:30 Tom^: yeah
09:31 Tom^: its like if the gpu sensor is wonky so it panics and downclocks for a second or two, meanwhile nouveau doesnt.
09:31 Tom^: *shrug
09:31 Tom^: *
09:31 karolherbst: most likely the case
09:31 karolherbst: or maybe not
09:31 karolherbst: who knows
09:32 karolherbst: mupuf: do you have a g84?
09:32 mupuf: yes
10:09 karolherbst: mupuf: do you have by any chance some SLI cables?
10:34 Tom^: speaking of random things, are you guys running something like this? https://i.pinimg.com/originals/d0/eb/1a/d0eb1aa68b0d1d6415bbf64d9ae78402.jpg a wall mounted uh pc! think of the ease of swapping cards and dust cleaning :D
10:37 RSpliet: I wouldn't want to rely on PCI-E flexicables. Signal integrity might suffer
10:39 Tom^: you could just stick it into the motherboard instead then heh, but yeah. would be kind of cool actually doing something like this
10:39 Tom^: then having that big box always in the way on the floor
16:26 karolherbst: airlied: as it seems, the solution to my problem was pretty simple: https://github.com/karolherbst/nouveau/commit/55e0e5e0326571fe3e102faaa364a8f5b316c1c5
16:27 karolherbst: nvkm_clk_pwrsrc is called on an event which isn't triggered when the device is suspended, pretty neat
16:34 karolherbst: ohh wait, it isn't syncing, bleh
16:35 karolherbst: okay, after a second though, it shouldn't be needed
18:49 anEpiov: has opencl advanced since I last asked?
18:49 anEpiov: also I've come accross a curious error
18:49 anEpiov: 'cannot stat /usr/bin/shaders'
18:50 anEpiov: which program got that folder?
18:51 anEpiov: ppsspp got an instant death
18:58 imirkin_: anEpiov: no status change on opencl.
19:22 anEpiov: GLES.cpp link shader
19:22 anEpiov: hard crash
23:36 Lyude: Anyone know if forcing on DisplayPort connections with nouveau (via echo connected > /sys/class/drm/card*-DP-*/status) is known to work?
23:36 Lyude: specifically for a monitor that for reasons that have been fixed in a later kernel release, cannot read the display's EDID
23:46 airlied: Lyude: forciing on displayport isn't really possible, since you have to link train and succeed at it
23:48 karolherbst: qdel: you can't reclock while the GPU is suspended, but I would like to dig into your reclocking issues if you can provide more detailed information about your issue
23:49 karolherbst: Lyude: pro tip: disable the EDID checksuming, some displays have wrong checksums
23:50 karolherbst: I know there was a display with a valid checksum on DVI, but a broken one on HDMI, because one byte was different in the EDID, but the checksum stayed the same
23:54 karolherbst: allthough it would be odd, because I don't think that the edid wasn't validated at some point. Lyude maybe other ways of connecting that display are working? I am just generally interested in that wrong EDID checksum situation, because I would like to know how common that is