12:05dcomp_: finally was able to get nvidia compiled and an mmiotrace attached to https://bugs.freedesktop.org/show_bug.cgi?id=95188
12:44Dezponia: karolherbst: Hai. Sorry I didn't turn up last weekend but if you have time to help me through testing your code on the TITAN Black today I'm up for it :)
12:46karolherbst: Dezponia: well just compile from the branch and install it :)
12:53karolherbst: hakzsam_: by the way, the broken tressfx rendering is unrelated to the other rendering issue :/ though with your align patch it doesn't crash except at start
13:01Dezponia: karolherbst: Not sure which branch to get :) I'm very much a user with eccentric tastes rather than a developer so some hand holding will be required (again :P)
13:06Dezponia: Yay I found it! I'm like a master haxxor for finally figuring out github and also finding your page there :P
13:06Dezponia: I dont get why I'm not in charge of the kernel already by now
13:07Dezponia: karolherbst: cloning
13:08Dezponia: Actually no i think I just cloned the wrong thingie
13:14Dezponia: cant compile, complains about missing a fle during the compile
13:14Dezponia: karolherbst: ./drm/nouveau/uapi/drm/nouveau_drm.h:30:21
13:15karolherbst: Dezponia: you have compile inside drm
13:17Dezponia: karolherbst: Not following, sorry :P
13:17karolherbst: Dezponia: there is a folder drm
13:18Dezponia: karolherbst: Yepp and there is a makefile in it but that fails as well :)
13:19karolherbst: yeah but it should have a different error
13:20Dezponia: karolherbst: "/nouveau/nouveau-stable_reclocking_kepler_v5/drm/nouveau/include/nvif/os.h:34:28: ödesdigert fel: soc/tegra/fuse.h:" again the file is missing etc
13:21karolherbst: yeah, you can remove that soc include
13:21Dezponia: I remember doing something similar last time :)
13:23karolherbst: mupuf: okay, I can confirm that all three channels have to be read out :/
13:23karolherbst: mupuf: but I don't know why we have to :/
13:24Dezponia: Yay stuffs building!
13:30Dezponia: karolherbst: the drm compile finished
13:31Dezponia: Not sure it worked though, seems to have just stopped on "stage 2"
13:46karolherbst: Dezponia: mhh, well the command should exit at some point
13:47Dezponia: karolherbst: pw titan: https://paste.kde.org/pbbdoiytz/msn1fi
14:14karolherbst: Dezponia: looks good
14:25karolherbst: mupuf: no idea :/ maybe nvidia polls just all channels and checks if it gets anything and just reads it out :/
14:27karolherbst: mupuf: ahh the rails in the vbios contain some information about how to configure the extdev.. somewhat
14:33Dezponia: karolherbst: Sooo... what do I do after that then?
14:33karolherbst: Dezponia: overwrite your nouveau module inside /lib/modules
14:36karolherbst: mupuf: :O the extedv config is in the rail :O
14:36karolherbst: byte 0a and 0b; 40 07
14:36karolherbst: reg 00:0x4e07
14:37karolherbst: I meant 0b, 0c in the rail: 07 4e
14:56karolherbst: stupid nvidia :/
14:56pecisk: karolherbst: what's up?
14:57karolherbst: nvidia is being weird and the power sensor too
15:01karolherbst: what the hell
15:01pecisk: sounds like discovery
15:01pecisk: ok, doesn't sound good
15:01karolherbst: when the channels on the ina3221 are disbaled, the old values just stay...
15:01pecisk: what's best way to get custom Mesa on let's say Fedora? Just build it?
15:03karolherbst: right, stupid sensor
15:37karolherbst: mupuf: also the polling interval depends on the current power consumption, not on the perf level
15:44karolherbst: Dezponia: did everything work out?
15:48Dezponia: karolherbst: Got a bit distracted but no idea. copied it into /lib/modules and rebooted but no idea if thats the right thing to do or if its actually using the new driver. I'm on Arch kernel 4.5.1 BTW
15:49Dezponia: mesa 11.2.1
15:49karolherbst: Dezponia: you need to find the old nouveau.ko inside /lib/modules
15:50karolherbst: should be /lib/modules/$kerne_version_tag/kernel/drivers/gpu/drm/nouveau
15:51Dezponia: There I have a nouveau.ko.gz
15:51karolherbst: mupuf: this is stupid :/ I think the rail in the vbios actually contains three mohm values and three values in addition to that and I don't find what would change nvidia to poll less channels :/
15:51karolherbst: Dezponia: yeah, remove it and move the compiled nouveau.ko inside that directory
15:53Dezponia: karolherbst: done
15:56karolherbst: then reboot :)
15:56Dezponia: karolherbst: Done and didn't work :P Not getting any graphics :)
15:58karolherbst: mhhh any error messages?
15:58karolherbst: I hope you didn't name it nouveau.ko.gz ;)
15:58Dezponia: I didn't :P
15:58karolherbst: mhh then full dmesg output pls
15:59Dezponia: Anyway I booted into my grsecurity kernel, didn't mess with the driver there
15:59karolherbst: well without the dmesg I can hardly tell what wen wrong
16:00Dezponia: karolherbst: Problem was that I couldn't switch to another tty
16:00Dezponia: The screen just started blinking when i did
16:01Dezponia: Blinked back and forth between the init tty and whatever I switched to about once every half second or something
16:01Dezponia: Trying again
16:01karolherbst: there should be no change without changing the pstates
16:02karolherbst: well you should be able to ssh into the machine
16:02Dezponia: karolherbst: True. I'll set that up
16:15Dezponia: karolherbst: http://pastebin.com/ycr1tLHZ
16:16karolherbst: Dezponia: do modprobe nouveau in ssh
16:18Dezponia: karolherbst: modprobe: ERROR: could not insert 'nouveau': Unknown symbol in module, or unknown parameter (see dmesg)
16:18karolherbst: then check dmesg ;)
16:18Dezponia: Nothing new
16:18Dezponia: just sddm segfaulting
16:19karolherbst: then insmod
16:19karolherbst: and pass the moudle path
16:19karolherbst: but you may have to modprobe ttm and some other modules first :/
16:19karolherbst: but this is odd
16:20karolherbst: I know there are sometimes random errors like this, but that shouldn't happen
16:20Dezponia: karolherbst: Worked, screen lit up with sddm and I can login to KDE
16:20karolherbst: wtf modprobe
16:21Dezponia: Exciting stuff
16:22karolherbst: in debugfs /sys/kernel/debug/dri/0 are two files
16:22karolherbst: "boost" and "pstate"
16:22karolherbst: you should be able to change to 0f pstate without problems
16:23karolherbst: and after that I would like to know your output of sensors
16:29Dezponia: karolherbst: http://pastebin.com/xYeW9KG5
16:29Dezponia: fan slowly climbing. Now 2040rpm at 58C
16:30karolherbst: yeah, the power sensor stuff is also not quite right on your
16:30Dezponia: Pfff 42W seems accurate, I got the TOYOTA PRIUS version of the GK110
16:30karolherbst: it is too low
16:31Dezponia: 2100rpm 59C
16:31karolherbst: it should be more around 90W
16:31karolherbst: a normal titan already consumes around 70W on 0f idling
16:31karolherbst: on nvidia
16:32karolherbst: Dezponia: you could do a unigine heaven benchmark on 8xmsaa ultra/normal tess on boost 2
16:32karolherbst: just watch over the temperature a bit
16:32karolherbst: I am sure it will reach like 80°C or something
16:34Dezponia: karolherbst: I'm getting artifacting on KDE 5.6.x BTW :) Reclocked or not
16:34karolherbst: even with stock nouveau?
16:35Dezponia: karolherbst: Only some slight with stock nouveau I think. Didn't test that a lot today
16:35karolherbst: ahh okay
16:35Dezponia: Running heaven now
16:36karolherbst: did you change the boost level?
16:36Dezponia: Only thing I know i saw on stock was some slight "rainbow" effect when using plasma widgets to monitor my CPU load. At the bottom of the graph there was a pixel line of rainbow stuff
16:36Dezponia: karolherbst: I did not change the boost right now, just pstate
16:37Dezponia: Want to see how it compares against my old heaven score with highest pstate :)
16:37Dezponia: 3000rpm 68C
16:37karolherbst: well it will be less
16:37karolherbst: because it is clocked to 888MHz or something
16:38Dezponia: karolherbst: What was the command to check on the clock BTW?
16:38karolherbst: cat pstate
16:38karolherbst: last line
16:39Dezponia: So you guessed correctly :)
16:39Dezponia: Or "guessed
16:39karolherbst: cat boost
16:39karolherbst: then you know how I knew it
16:40Dezponia: Seems rock solid on 68C at 3000RPM
16:40Dezponia: So thats neat at least
16:40Dezponia: Using 60W! Damn my GPU is so efficent! Its totally magic and real and not at all a bug
16:40karolherbst: the power reading is wrong
16:40karolherbst: I could cook up a patch though :D
16:41Dezponia: By all means, something you need from my side?
16:41Dezponia: But dont waste time unless its actually useful :) I dont NEED to know how much power its using :)
16:41Dezponia: Also Heaven seems to render fine so far, no articats I can spot
16:41karolherbst: no, we have a titan now anyway
16:41karolherbst: Dezponia: 8xmsaa?
16:41Dezponia: karolherbst: Yepp
16:42karolherbst: I guess your old test used 4xmsaa
16:42karolherbst: because 8xmsaa was broken back then
16:42Dezponia: karolherbst: Yepp, 4x. I saved the benchmark result file
16:44Dezponia: karolherbst: Changed boost to 2
16:44karolherbst: now it is getting hot :D
16:44Dezponia: Core 1072 memory 6999
16:45Dezponia: Going to start heaven again then
16:45karolherbst: Dezponia: well
16:45karolherbst: Dezponia: you should be able to see the core clock dropping whiel the temperature rises :D
16:45karolherbst: maybe not
16:46Dezponia: Only if it actually goes high enough. Didn't drop ever on your old patches so no idea why it would now
16:46karolherbst: because it wasn't implemented in the old ones
16:47Dezponia: karolherbst: We'll see then
16:47karolherbst: and with drop I mean a drop by 23MHz or something
16:47Dezponia: claims about 70W usage, 3750RPM 74C
16:47Dezponia: 1059 core 6999 mem
16:48Dezponia: So seems to be dropping properly
16:48Dezponia: karolherbst: At what temprature should it be dropping?
16:48karolherbst: Dezponia: no idea
16:49Dezponia: 3780RPM tops I've seen and never above 74C
16:49karolherbst: didn't looked in detail into your vbios
16:49Dezponia: karolherbst: BTW you said you had a TITAN now, someone on the team got one? :)
16:49karolherbst: kind of
16:49karolherbst: but nvidia-smi prints the power consumption
16:50karolherbst: which gives us the possibility to RE those power sensor stuff right
16:50Dezponia: karolherbst: I have not seen it drop under 1059 yet either BTW
16:51karolherbst: Dezponia: well if you are done with heaven you could nvaforcetemp 1 and then it should raise
16:52Dezponia: karolherbst: I'll happily provide that RE stuff from nvidia-smi btw if it would be useful for you
16:53karolherbst: well we can do it ourself ;)
16:53karolherbst: just takes some time
16:53Dezponia: no differences between TITAN and TITAN Black?
16:54karolherbst: not really
16:55Dezponia: GPU core volt is 1.15 in Heaven BTW
16:57Dezponia: And it jumped back to 1072MHz after closing heaven and reaching 66C. Core volt at 1.19
16:58Dezponia: Going to try at 4x MSAA just for a quick comparison to the old benchmark file :P
16:58Dezponia: See if its magically going to be faster
17:00Dezponia: karolherbst: Anything else that would be useful to run or test while we're at it?
17:00karolherbst: Dezponia: not really
17:00Dezponia: Will test that nvaforcetemp 1 thing once this heaven run is done
17:00karolherbst: Dezponia: besides stability
17:00karolherbst: Dezponia: no, it is actually fine now
17:01karolherbst: your clock increased at 66°C ;)
17:01Dezponia: Well thats my guess at any rate. Sensors said it went back up after I hit that
17:02Dezponia: Now you guys just need to implement quad SLI with 100% scaling per card and I'll get 3 more TITAN Blacks and everything will be perfect. Easy
17:03karolherbst: I think I will clean my laptop now anyway :D
17:03karolherbst: Dezponia: how much more perf do you get with the newest patches?
17:04Dezponia: Seems I got into the ballpark of that old benchmark score. like 22 points lower or so on the first run. Not really bad since the clock is lower (now that it actually downclocks properly with temprature)
17:04Dezponia: Doing a second run for fun.
17:04karolherbst: but you used that link.max hack before, right?
17:04karolherbst: ohh wait
17:05Dezponia: No idea, some patches from you :P Built a 4.3 kernel with latest mesa and your patches last time
17:05karolherbst: even with that you got higher clocks, because the voltage wasn't limited by temperature
17:05karolherbst: ah well
18:24karolherbst: cleaned my laptop, 6°C cooler :/
18:25karolherbst: Dezponia: how is it going until now?
18:26Dezponia: karolherbst: Good but the same KDE issues remain when running desktop effects. Switching off desktop effects removes any artifacting i could see. I had no issues in Heaven etc
18:26Dezponia: And the little bit of youtube I tried worked
18:26karolherbst: Dezponia: version of mesa?
18:26Dezponia: mesa 11.2.1
18:27karolherbst: my fans are doing crazy sounds :/
18:28Dezponia: karolherbst: https://seafile.dezponia.se/d/54e02cc0a3/
18:28Dezponia: Benchmark results BTW
18:29karolherbst: boost 0/1/2 ?
18:30Dezponia: Nah, 0 2 and 2 with 4x
18:30karolherbst: yeah I noticed it yet
18:30Dezponia: Compared with the best old run I had of 1103
18:31Dezponia: So seems fairly accurate considering the core is like 25MHz lower
18:31karolherbst: nothing we can do about thought :D
18:31karolherbst: we might add some OC features later
18:31Dezponia: Also didn't re-run a lot to see what I could push out trying to get an optimal score. Just 2 runs or something
18:32karolherbst: yeah, it is fine
18:32Dezponia: That 1103 score took a while to get :)
18:32karolherbst: there is a thing though
18:32karolherbst: there is a config=NvVoltOffsetmV option
18:32karolherbst: the original idea was to increase the voltage if the voltage set by nouveau is too low or causes instabilities
18:32karolherbst: it works the other way around too
18:33karolherbst: it keeps some rules in mind though, so the best what you can do is usually something around +-150mv
18:33karolherbst: but if you load nouveau with config=NvVoltOffsetmV=-100mv
18:33karolherbst: nouveau will calculate a 100mv lower voltage for each cstates
18:33karolherbst: which may or may not enable higher clocks at the cost of stability
18:34Dezponia: No worries :) Any idea when you guys will be comfortable with enabling auto-reclocking?
18:37Dezponia: BTW a friend of mine just got their hands on a GTX 980 SuperClocked from EVGA. If you need some tests run for that card I could probably convince them to run a few
18:46karolherbst: Dezponia: well there is no memory reclocking on those yet
18:46Yoshimo: a vbios for comparison maaybe?
18:47Dezponia: Cant hurt to ask :3
18:51karolherbst: Yoshimo: nope, we need the PMU firmware
18:52Yoshimo: so there is no difference between the vbios files of the same model from different manufacturers?
18:53karolherbst: well not for the titans
18:53karolherbst: I think
18:54karolherbst: and the vbios doesn't really help that much with things
18:54karolherbst: only if something is different
18:54karolherbst: and this can happen on everygpu
20:06imirkin: skeggsb: have a look at https://bugs.freedesktop.org/show_bug.cgi?id=95188 -- maybe there's a reset sequence necessary for some mobile GM108's just like you had to do for GK10x?
20:06imirkin: skeggsb: mmiotrace included and all.
20:34karolherbst: yay, finally rpm data for my fans :)
21:56karolherbst: do you think it might make sense to add code in nouveau to control external fans? Like a wmi module could export GPU fans and nouveau could control them (if write functions are also there)
22:05imirkin: EC normally controls those no?
22:07karolherbst: but at least on mine, I can also control them manucally through the EC
22:11karolherbst: the thing is, my EC is a stupid piece of crap setting dutys to 50% at like 55°C :/
22:11karolherbst: and on 45°C the duty is like 0
22:16imirkin: but it shouldn't be up to nouveau to control that...
22:21karolherbst: I know
22:21karolherbst: still annoys me though :/
22:21karolherbst: maybe I can change the thresholds in the EC