10:16kreyren: Can i use nouveau to get me raytracing on a secondary GPU?
10:19HdkR: Maybe compute shader based RT
10:20HdkR: Write a ray tracer, port it to GLSL :)
10:20karolherbst: don't give them ideas
10:21kreyren: o.o
10:21kreyren:was thinking about something done in RISC-V originally
10:22karolherbst: well, then you need vulkan or GL 4.6
10:22HdkR: Why? Compute shaders are in GL 3.x :)
10:22karolherbst: not if you want to use spir-v
10:23HdkR: They said riscv though
10:23karolherbst: ohhhh
10:23karolherbst: my mistake then
10:23kreyren: Or like is there a better way to get raytracing using libre software/hardware ?
10:23kreyren: afaik vulkan doesn't do raytracing?
10:23kreyren: neither does GL? O.o
10:23karolherbst: AMD + vulkan
10:23kreyren: well they do but poorly?
10:24HdkR: Just write it in C and run it on a CPU
10:24karolherbst: :D
10:24kreyren: O.o
10:24kreyren: <HdkR "Just write it in C and run it on"> would that actually work
10:24HdkR: Of course
10:24karolherbst: C is turing complete
10:24karolherbst: and so is MS excel
10:24damo22: write it in TeX
10:24karolherbst: is tex turing complete?
10:24damo22: i believe so
10:25HdkR: Don't let the hype of GPU vendors having some hardware to accelerate RT. You don't need it to play in RT
10:25kreyren: like i meant would it be sufficiently efficient for the use in gaming?
10:25kreyren: seems that AMD+vulkan is too slow compared to nvidia's built-in ?
10:26HdkR: Start with a small scope. You need to support RT before thinking about gluing it in to a game
10:26damo22: fast inverse sqrt
10:26damo22: ftw
10:30kreyren: What's RT
10:30kreyren: ah raytracing
10:30kreyren: hmm O.o
10:31kreyren: thanks for brainstorm
13:48iblue: Hello. I have the issue that my GPU fans are not spinning. Actually they stay at the speed they are, when the system was booting. When booting with 0 RPM, they stay at 0, not matter the temperature. I have dual GTX 1080, nouveau says in the kernel log it is version 1.3.1 (which I don't believe) and pwm1_enable returns -1 on both cards, writing to
13:48iblue: it does not work. Kernel is 5.10.8. Any suggestions?
13:51karolherbst: iblue: huh...
13:51karolherbst: could be that we do the inverse
13:52karolherbst: iblue: might booting with nouveau.debug=therm=trace?
13:53iblue: I will try. Hold on a sec.
13:57iblue: The parameter is correct? No quotes?
13:59iblue: Anyway. Will see. Rebooting now.
14:14iblue: karolherbst: I get this in my dmesg now:
14:15iblue: [ 3.358603] nouveau 0000:08:00.0: therm: preinit running...
14:15iblue: [ 3.358607] nouveau 0000:08:00.0: therm: preinit completed in 0us
14:15iblue: [ 3.358629] nouveau 0000:08:00.0: therm: fini running...
14:15iblue: [ 3.358631] nouveau 0000:08:00.0: therm: fini completed in 0us
14:15iblue: [ 3.390217] nouveau 0000:08:00.0: therm: init running...
14:15iblue: [ 3.390221] nouveau 0000:08:00.0: therm: one-time init running...
14:15iblue: [ 3.390228] nouveau 0000:08:00.0: therm: FAN control: toggle
14:15iblue: [ 3.390233] nouveau 0000:08:00.0: therm: internal sensor: yes
14:15iblue: [ 3.390235] nouveau 0000:08:00.0: therm: one-time init completed in 12us
14:15iblue: [ 3.390238] nouveau 0000:08:00.0: therm: programmed thresholds [ 90(3), 95(3), 105(5), 135(5) ]
14:15iblue: [ 3.390254] nouveau 0000:08:00.0: therm: init completed in 33us
14:15iblue: [ 4.832954] nouveau 0000:09:00.0: therm: preinit running...
14:15iblue: [ 4.832960] nouveau 0000:09:00.0: therm: preinit completed in 0us
14:15iblue: [ 4.855220] nouveau 0000:09:00.0: therm: fini running...
14:15iblue: [ 4.855225] nouveau 0000:09:00.0: therm: fini completed in 0us
14:15iblue: [ 5.212939] nouveau 0000:09:00.0: therm: init running...
14:15iblue: [ 5.212946] nouveau 0000:09:00.0: therm: one-time init running...
14:15iblue: [ 5.212955] nouveau 0000:09:00.0: therm: FAN control: toggle
14:16karolherbst: mhhhh
14:17karolherbst: mupuf: how does the toggle thingy work?
14:17RSpliet: iblue: you may want to post the full unfiltered log to a paste website and share the link here
14:17karolherbst: iblue: do you see any fan updates in the log?
14:17karolherbst: usually we have regular checks with appropiate updates
14:18karolherbst: iblue: what was the hottest the GPU went to?
14:18karolherbst: I would assume we just configure something incorrectly
14:19karolherbst: iblue: mind creating a full bug at https://gitlab.freedesktop.org/drm/nouveau/-/issues with your current dmesg attached + vbios.rom from your affected GPU?
14:19iblue: I didn't measure it, but the back of my case got so hot, I couldn't touch it. I guess 80? When I opened it, it smelled a little like burned PCBs.
14:20karolherbst: vbios.rom can be found at /sys/kernel/debug/dri/*/vbios.rom but I guess it's enough to use the one from the first GPU
14:20karolherbst: iblue: "sensors" will report it
14:20iblue: Currently it says 53C and 0 RPM.
14:20iblue: Last boot it was 49C and 700RPM.
14:21iblue: I think it is stuck on the speed it booted in.
14:21karolherbst: could be
14:21karolherbst: anyway, filing a bug with the information might help
14:21iblue: Ok, will do. Thanks.
14:23RSpliet: iblue: also, have you tried other kernels at all? Has this set-up worked for you previously on nouveau?
14:24RSpliet: I don't think this code has changed at all, but could be useful info nonetheless
14:24RSpliet: karolherbst: should this work when firmware upload fails on gtx1080?
14:24iblue: No, I have not. I was usually running the proprietary driver and it worked well. I had to set CoolBits in the xorg and control the speed manually, not because it didn't work, but because the Fans start and stop, which make a snorring sound, so I keept them at 20%.
14:24karolherbst: uhm............
14:25karolherbst: ahh crap
14:25karolherbst: that's pascal
14:25iblue: I thought, it may be, because the GPU is stuck in this manual mode, so I booted the proprietary drivers again with auto fan control, then booted into nouveau. No changes.
14:25karolherbst: iblue: yeah... that's a driver thing anyway
14:25karolherbst: sooo
14:26karolherbst: the deal is, on pascal we require signed firmware to actually change the fan speeds, _but_ we even fail to detect the proper fan type on your GPU as it seems
14:26karolherbst: it's.... a stupid issue
14:27karolherbst: I would take a look at your vbios and see if there is some magic we can do on your GPU, but usually we simply can't control the fans
14:27karolherbst: not that we don't know how, the hardware doesn't let us
14:27iblue: Yes, I understand.
14:28imirkin: damo22: the real key is to write it in postscript. then you can just run it directly on your printer.
14:28RSpliet: iblue: this is part of the reason why i asked you to paste an unfiltered kernel log and share the link here. It gives more hints. Anyway, if you're writing a bug report, just stick it there instead, it's fine
14:28karolherbst: iblue: I'll take a look at your vbios if there is anything special, but no promises that it will help
14:28iblue: Ok, will do that. Can I grep for nouveau or do you need the full log?
14:29imirkin: iblue: someone else came in here recently with a seeming therm management issue on a later-model GPU (i.e. where we can't control it)
14:29karolherbst: uhm... greping for nouveau would be fine, but generally there is no sensetive information in dmesg
14:29imirkin: i don't think there's anything we can do about it
14:29karolherbst: imirkin: I at least want to know about all the different vbios configs :D
14:30imirkin: karolherbst: ok, search up... bamse had an issue like that, perhaps you can ask him for more details
14:31karolherbst: yeah.... seems like this week is just as crazy as the last two ones and I have to deal with other stupid bugs :)
14:31RSpliet: iblue: I'd recommend not grepping. It's good to also have non-nouveau data in there (kernel version/command line, weird other drivers that may fault or taint, non-nouveau DRM messages etc.)
14:32RSpliet: If the reader wants to filter they can grep themselves
14:32karolherbst: RSpliet: we should have an issue template :D
14:32karolherbst: RSpliet: do you have time to create one?
14:32iblue: Ok, then it will take a while, I will redact some things.
14:32RSpliet: karolherbst: https://nouveau.freedesktop.org/Bugs.html
14:32RSpliet: Knock yourself out
14:33karolherbst: RSpliet: yeah, but I meant in gitlab
14:33karolherbst: you can have an issue template there
14:33RSpliet: Oh, right. I don't have time
14:33karolherbst: :D
14:34RSpliet: I mean, that's not nearly as passive-agressive as it sounded. I've been exhausted since forever, doing a full-time job and getting a paper published
14:34karolherbst: that's like the same thing you say for 5 years :p
14:35RSpliet: Yes. Luckily I finally have the money now to fix some of my physical health issues (bad back and posture)
14:35bamse: imirkin: don't have any more details :( i can read the fan speed, but the driver prevents me from switching to manual or auto controls
14:35RSpliet: Just not the places to spend it and get it solved, thanks to a pandemic.
14:35imirkin: bamse: i think karolherbst wanted your vbios? and also my recollection of your issue is ... fleeting
14:36karolherbst: RSpliet: like buying a proper desk and chair for 1.5k total? :p
14:36imirkin: bamse: or was the issue merely the lack of manual control?
14:36imirkin: i thought you were seeing higher temps. or higher fan speeds. or something
14:37bamse: imirkin: my problem is that something before linux boots sets the fan speed, and then it never changes
14:37iblue: bamse: Same here
14:37bamse: imirkin: and attempting to set the hwmon controls to manual or auto is -EINVAL
14:37karolherbst: imirkin: so there is this magic PMU image in the vbios which can do such crazy things
14:37karolherbst: :D
14:37imirkin: bamse: yeah, that's expected - there's no manual control.
14:38bamse: imirkin: okay, and the comment in the code says that my pmu should magically take care of it
14:38karolherbst: I just don't know on what base
14:38imirkin: bamse: yes, it should.
14:38karolherbst: you know what? I might hack around with it and see if we can do crazy magic
14:38bamse: imirkin: right, makes sense to me as well...but them fans aren't spinning
14:39RSpliet: karolherbst: and a proper mattress, and supportive soles yes
14:39karolherbst: might have to do it anyway
14:39karolherbst: RSpliet: ahh, yeah
14:39karolherbst: money well spent
14:39karolherbst: I bought all of that after moving to germany as well
14:39karolherbst: if you need pointers
14:39RSpliet: as soon as mattress stores open
14:39karolherbst: RSpliet: just order them
14:40RSpliet: Can't. I need a "special size" they sell in NL but not UK. And Brexit
14:40karolherbst: ehhhh uffff
14:40RSpliet:== 2 metres
14:40karolherbst: shit
14:40karolherbst: at least I could help with the chair :p
14:40karolherbst: my desk ist from a german store, so that wouldn't help either
14:41karolherbst: but with everything, electric motor, cable management shit
14:41RSpliet: "ist" :-D Nah I got an electric sit/stand desk already, "relocation expense"
14:41karolherbst: ahh
14:41karolherbst: cool
14:41RSpliet: Chair as well, plus extended gas lift. Because 2 metres
14:41karolherbst: in Germany that's all tax deductible
14:41karolherbst: so I get like 35% back
14:41RSpliet: Yeah here as well, I didn't pay a penny for those
14:41karolherbst: nice
14:42RSpliet: Relocation benefit when I joined the company
14:42karolherbst: I bought them from my own money though
14:42karolherbst: ahh yeah
14:42RSpliet: Also, I ordered a monitor with a VESA arm back in January
14:42karolherbst: I only got tax advisors for free
14:42karolherbst: yay
14:42karolherbst: me as well
14:42RSpliet: still hasn't arrived, due for the 20th of April now. Because Brexit
14:42karolherbst: :/
14:43karolherbst: ehhh crap
14:43karolherbst: gitlab issues have to be files in the repository
14:43karolherbst: that's... annoying
14:43karolherbst: I mean templates
14:44karolherbst: I have to trick the system then
15:03iblue: So here is the issue: https://gitlab.freedesktop.org/drm/nouveau/-/issues/75
15:05karolherbst: cool thanks
15:05iblue: Immer gerne.
15:13iblue: Hey, that might be a stupid question, but what's the development process for those drivers? You have two cards and attach one to a Xen VM, so you don't have to reboot and a working screen all the time?
15:25karolherbst: ssh helps
15:26karolherbst: but yeah, kind of depends
15:26karolherbst: you can just work on the same machine and have to reboot
15:26karolherbst: but you don't have to touch the kernel for everything
15:28iblue: Ok, I see. Thanks for the help so far.
15:29RSpliet: karolherbst: Silly idea, but we had a fix to PMU reset stuff after that person reported timeouts. Could that same fix help with keeping the PMU in charge of the fans? or is that just complete silliness? :-D
15:30RSpliet: detecting the fan as a toggle fan sounds dodgy too though...
15:49imirkin: doesnt matter, we still can't control it...
15:54karolherbst: wellllll
15:54karolherbst: imirkin: maybe we don't have to
15:54karolherbst: so the PMU is for the real basics
15:55karolherbst: _maybe_ it just sets it based on the current time
15:55karolherbst: *temp
15:55karolherbst: and if we just reboot it it might update it
15:55imirkin: not to mention phase of the moon ;)
15:55karolherbst: of coutrse