03:28 degasus: "codegen/nv50_ir_ra.cpp:1330:29: error: 'isinf' was not declared in this scope", a missing std:: ?
03:31 degasus: nvm, found the thread on mesa-dev
03:38 SaveTheRobots: Tom^: you getting any instability with your 780 Ti while running in pstate 0f and boost 1? i can managed a whole 3 mins of L4D2 before it locks up :(
03:39 SaveTheRobots: manage*
04:57 SaveTheRobots: karolherbst: my machine locks up after around 3-4mins of gaming using pstate 0f and boost level 1 :(
04:57 karolherbst: SaveTheRobots: k, do you have the dmesg output from that time?
04:58 karolherbst: and then I would also need your vbios.rom
04:59 SaveTheRobots: there's nothing in my logs from the last lockup
04:59 SaveTheRobots: and shall i use these instructions to dump my vbios? https://nouveau.freedesktop.org/wiki/DumpingVideoBios/
05:01 karolherbst: SaveTheRobots: you should have a file called vbios.rom where the pstate file is
05:01 karolherbst: you can cat it and save the output somewhere
05:03 SaveTheRobots: https://transfer.sh/Yo0tP/vbios.rom.gz
05:04 karolherbst: looks good
05:05 SaveTheRobots: :D
05:05 SaveTheRobots: what do you think caused it then? clock speeds too high?
05:05 karolherbst: currently checking
05:06 SaveTheRobots: ah, thanks
05:07 karolherbst: how difficult would it be for you to run the nvidia driver?
05:07 SaveTheRobots: easy, what d'ya need?
05:08 karolherbst: I wrote a tool which checks if nouveau sets the right voltage
05:08 karolherbst: but for this, you have to run the nvidia driver
05:08 SaveTheRobots: yeah, no worries, hit me :)
05:09 karolherbst: in the nouveau branch of mine, you can run make in the top level dir and then start this tool like this: LD_LIBRARY_PATH=lib bin/nv_cmp_volt
05:11 SaveTheRobots: so, install and boot to NVIDIA drivers, run this tool and load a game while it's running?
05:11 karolherbst: right
05:11 karolherbst: it will print out some stuff
05:12 SaveTheRobots: cool, i'll go do it and pastebin the resulting output
05:13 karolherbst: SaveTheRobots: and I also need which clock nvidia sets
05:13 karolherbst: you can see this in nvidia-settings
05:13 karolherbst: this tool might read out some clocks wrong, so the output is a bit useless without the clock set
05:20 SaveTheRobots: in the PowerMizer tab?
05:21 karolherbst: yeah
05:21 karolherbst: there are some ranges, but also the currently set clocks
05:22 SaveTheRobots: CC lib/tegra.o
05:22 SaveTheRobots: In file included from lib/drm.c:37:0:
05:22 SaveTheRobots: ./drm/nouveau/uapi/drm/nouveau_drm.h:30:21: fatal error: drm/drm.h: No such file or directory
05:22 SaveTheRobots: compilation terminated.
05:22 SaveTheRobots: when compiling in top level nouveau repo
05:22 karolherbst: you can add a // before that include
05:23 SaveTheRobots: k
05:23 karolherbst: not quite sure why that happens, it isn't important for the tools though
05:24 SaveTheRobots: ah :p
05:24 SaveTheRobots: ok, ready to go, do you need nvidia-settings clock readouts whilst gaming?
05:24 SaveTheRobots: or idle?
05:32 SaveTheRobots: nv_cmp_volt output: http://dpaste.com/2BBTNQN
05:32 SaveTheRobots: this was a few minutes of gameplay, alt-tabbing out a few times
05:33 SaveTheRobots: nvidia-settings while gaming: https://imgur.com/pxyA4mo
05:42 SaveTheRobots: at least i think that nvidia-settings screenshot is accurate, i had to alt-tab out to get it, so it might have clocked down while i was taking the screenshot?
05:49 karolherbst: SaveTheRobots: you should also run the game with vsync disabled
05:49 karolherbst: otherwise you might not hit the max clocks
05:58 SaveTheRobots: hmm, i believe it was disabled, but i'll check
05:59 SaveTheRobots: ah, it wasn't, i'll try again
06:16 SaveTheRobots: karolherbst: http://dpaste.com/21JY33Q / https://imgur.com/ZEJGBU2
06:17 SaveTheRobots: got a lot of stuttering/stalling during that playthrough.. that's new :\
06:26 SaveTheRobots: hmm nvm, looks like the stuttering was control-related, enabling raw input helped
06:49 karolherbst: SaveTheRobots: yep, that looks better
06:53 karolherbst: ohh I see something odd already
06:54 karolherbst: SaveTheRobots: can you boot with nouveau again?
06:56 SaveTheRobots: karolherbst: yup, booted now
06:57 karolherbst: k then clock to 0f and give me the voltage sensors reports
06:57 SaveTheRobots: http://dpaste.com/3715SEW
06:58 SaveTheRobots: GPU core: +1.05 V
06:58 SaveTheRobots: fan1: 1710 RPM
06:58 SaveTheRobots: temp1: +53.0???C
06:58 SaveTheRobots: 53C seems a little high? :O
06:58 SaveTheRobots: oh nvm
06:58 karolherbst: no why?
06:58 SaveTheRobots: nvm, talking crap, 0f explains that :p
06:59 karolherbst: as long as it stays below 95°C, everything is fine :D
06:59 karolherbst: I think the emergency shutdown of the gpu is around 105°C
06:59 karolherbst: 1.05V mhh
07:00 karolherbst: SaveTheRobots: 960MHz?
07:00 karolherbst: *980
07:00 karolherbst: last line in pstate
07:00 SaveTheRobots: 0f: core 549-1176 MHz memory 7000 MHz AC DC *
07:01 SaveTheRobots: AC: core 1046 MHz memory 6999 MHz
07:01 karolherbst: ohh right
07:01 karolherbst: I cap to the boost clock for now
07:01 karolherbst: not the base one
07:02 SaveTheRobots: shall i echo 0 > boost/
07:02 SaveTheRobots: ?
07:02 karolherbst: wow...
07:02 karolherbst: no
07:02 karolherbst: wait
07:03 karolherbst: mhh
07:03 karolherbst: the voltage indeed looks a bit low
07:04 karolherbst: but nouveau doesn't seem to volt too low
07:04 SaveTheRobots: so it's hopefully just a case of ramping up the juice? (without breaking my card :p)
07:04 SaveTheRobots: ah
07:05 SaveTheRobots: could i just limit boost to 0 instead?
07:06 karolherbst: usually this will be less stable
07:06 karolherbst: you could try to echo 2 into boost and see how that goes
07:06 karolherbst: but the gpu can crash for various reasons
07:07 SaveTheRobots: ok, is there a better solution then?
07:08 SaveTheRobots: i guess we'd need to know which freq the gpu is crashing at and give that freq more juice?
07:08 karolherbst: mhh
07:08 karolherbst: I just need the dmesg output when the crash happens
07:08 karolherbst: it could give us a good clue
07:09 SaveTheRobots: ok cool, how do i get dmesg output, seeing as it hard-locks ups ?
07:09 SaveTheRobots: up*
07:10 karolherbst: SaveTheRobots: try ssh
07:11 SaveTheRobots: k 1 mo, i'll boot my MBP up
07:11 SaveTheRobots: shall i test 0f/boost1 then ?
07:15 karolherbst: boost 1 is default
07:21 SaveTheRobots: karolherbst: http://dpaste.com/35W0NPQ
07:21 SaveTheRobots: repeated over and over
07:21 SaveTheRobots: is there anything else you need while it's crashed?
07:22 SaveTheRobots: [ 1678.184498] nouveau 0000:01:00.0: hl2_linux[4372]: failed to idle channel 11 [hl2_linux[4372]]
07:22 SaveTheRobots: [ 1680.189375] nouveau 0000:01:00.0: timeout at /home/john.slack/build/src/nouveau/drm/nouveau/nvkm/engine/fifo/gpfifogk104.c:48/gk104_fifo_gpfifo_kick()!
07:22 SaveTheRobots: [ 1680.189377] nouveau 0000:01:00.0: fifo: channel 11 [hl2_linux[4372]] kick timeout
07:22 SaveTheRobots: [ 1680.189381] nouveau: hl2_linux[4372]:00000000:0000a16f: detach gr failed, -16
07:22 SaveTheRobots: [ 1680.189390] nouveau 0000:01:00.0: fifo: SCHED_ERROR 0d []y
07:27 karolherbst: k
07:27 karolherbst: that looks like something which might be caused by too low voltage or something else in that area
07:27 karolherbst: we can try somehting
07:28 karolherbst: SaveTheRobots: but it only looks up after some time, right?
07:28 SaveTheRobots: correct
07:28 SaveTheRobots: trying boost 2 now for lulz
07:31 SaveTheRobots: no crashes yet on b2
07:32 SaveTheRobots: also, i'm guessing voltage is below what it should be because on NVIDIA while gaming, i get coil whine, no whine under nouveau at all
07:32 karolherbst: mhh
07:32 karolherbst: well the current voltage calculation isn't perfect yet
07:33 karolherbst: but it is still better than default
07:34 SaveTheRobots: definitely :p
07:34 SaveTheRobots: well, b2 actually seems stable
07:34 karolherbst: yeah, a lower voltage is a bigger prioblem on lower clocks
07:34 karolherbst: because the voltage error in nouveau is usually an absolute value and not a relative one
07:35 SaveTheRobots: ahh ok
07:35 karolherbst: but with b2 you have the problem that the gpu might draw too much power
07:35 SaveTheRobots: is it better we diagnose b1 then ?
07:35 SaveTheRobots: i'm happy to do whatever you need
07:35 karolherbst: mhh you can try out 0
07:35 karolherbst: and check if this crashes even faster
07:35 SaveTheRobots: sure, i'll set 0 while in game
07:38 SaveTheRobots: fan is a lot quieter, -10W power according to sensors
07:38 SaveTheRobots: no crashes yet
07:40 karolherbst: well nouveau actually also reads out the clock wrong on nvf0 cards and newer
07:40 karolherbst: but I doubt that causes any issues
07:40 karolherbst: SaveTheRobots: k, can you give me the clocks and voltage for the various boost levels?
07:40 SaveTheRobots: from pstate?
07:41 karolherbst: yeah
07:47 SaveTheRobots: karolherbst: http://dpaste.com/2CFNAWM
07:47 karolherbst: yeah, the boost1 voltage looks just too low
07:48 karolherbst: 68MHz more, but only 0.05V more
07:48 karolherbst: from 1->2 79MHz more, but 0,14V more
07:49 karolherbst: I would expect that the voltage should be around 0.05V higher
07:49 karolherbst: sadly we can't force nvidia to set a specific clock
07:50 karolherbst: and while nouveau reads them out false, the tool is a bit useless for this task
07:50 SaveTheRobots: ah, is this something i can modify myself in the source code or would you not recommend that ? i could increase in incremenets until it's stable
07:50 SaveTheRobots: ah :(
07:50 karolherbst: SaveTheRobots: well you can use boost 0 or 2 for now
07:50 karolherbst: 0 is super safe
07:50 karolherbst: but gives you 150Mhz lowe clocks
07:50 karolherbst: not that it matters much with your gpu
07:50 SaveTheRobots: :P
07:50 SaveTheRobots: yeah, 0 is still super fast and more than enough for my gaming needs
07:51 SaveTheRobots: but i'm happy to help if there's any testing/diagnosing you need doing
07:51 karolherbst: 0 also has the advantage that we might not exceed the power consumption limits (95% sure)
07:51 karolherbst: well 2 should be also save, but there might be ways to exceed the limit
07:51 SaveTheRobots: is 2 'safe' ? i'm in no danger of damaging my card right?
07:51 SaveTheRobots: ah o
07:51 karolherbst: no, the gpu will shutdown if something bad happens
07:51 SaveTheRobots: k
07:51 karolherbst: but
07:51 karolherbst: we didn't test yet what happens when we draw too much power
07:52 karolherbst: maybe your PSU gets upset
07:52 karolherbst: no idea
07:52 karolherbst: the three boost levels are in fact specific clocks the gpu "has" in the vbios
07:52 karolherbst: 0: base clock
07:52 karolherbst: 1: boost clock
07:52 karolherbst: 2: unlimited
07:52 SaveTheRobots: meh, i can just buy another, i assume if the PSU does crap out, it'll just crash, it won't explode and take out my other components?
07:52 karolherbst: 0 and 1 are the clocks GPUs are sold with
07:53 karolherbst: most of the cases they are caleld "base" and "boost"
07:53 karolherbst: but the "boost" clock is more like a clock the gpu will most likely _always_ reach while boosting
07:53 SaveTheRobots: 2 = max clock speeds then?
07:53 karolherbst: not really
07:53 karolherbst: more like what is currently possible
07:53 karolherbst: there are much higher clocks in the vbios
07:54 karolherbst: highest is 1307MHz
07:54 karolherbst: but this clock needs an absurd high voltage
07:54 karolherbst: so we can't set it
07:54 SaveTheRobots: ah ok, and probably not worth the boost anyway, considering the FPS increase i'd see (i imagine)
07:54 karolherbst: some lower clocks might be possible at a specific temperature, because depending on the temperate, you use a different voltage for the same clock
07:55 karolherbst: and also the temperate specifies the maximum volt we should set
07:55 karolherbst: so at 30°C a higher clock is possible than with 80°C
07:55 karolherbst: boost 2 tries to sets the gpu to the highest clock possible
07:55 karolherbst: nvidia usually does the same, but I am sure they also use the power consumption somewhat
07:55 karolherbst: nouveau doesn't
07:56 karolherbst: that's why I say 2 isn't "safe"
07:56 SaveTheRobots: ahh, i see
07:56 karolherbst: because it brings the gpu to it's limit actually (according to the vbios that is)
07:57 SaveTheRobots: the difference in fan speeds between 1>2 is incredible, it's almost a 800 RPM difference
07:58 SaveTheRobots: also, quick question, when unbinding nouveau from a vtconsole via echo 0 > /sys/class/vtconsole?/bind crashes the display and i have to reboot
07:58 karolherbst: yeah
07:58 SaveTheRobots: am i doing it wrong ?
07:59 karolherbst: SaveTheRobots: well I never done it with a display
07:59 karolherbst: only per remote
07:59 karolherbst: but I would say it disables all outputs of the gpu
07:59 karolherbst: you might be able to do that savely via ssh
07:59 SaveTheRobots: ahh ok
08:09 SaveTheRobots: karolherbst: you say that 0 and 1 are the clocks that the gpu is sold with, how can the voltage be too low? i guess because clocks/thermal profiles are in teh vbios but you're setting voltage manually?
08:16 karolherbst: SaveTheRobots: nouveau just doesn't parse the voltage table right
08:17 karolherbst: there is a voltage entry for each clock
08:17 SaveTheRobots: is this where the reverse engineering is coming in?
08:18 karolherbst: for the boost 1 clock we have those entries (they can be linked together): c0 15315536 c1 -2773 c2 -699 c3 0 c4 0 c5 0 + c0 -5580 c1 0 c2 34 c3 0 c4 0 c5 -124 + c0 500000 c1 0 c2 0 c3 0 c4 0 c5 0 + c0 0 c1 0 c2 0 c3 0 c4 0 c5 0
08:18 karolherbst: yeah
08:18 karolherbst: and out of those coefficients we have to calculate the right voltage
08:19 karolherbst: and the factors with what we have to multiply with is most likely only known to the nvidia driver
08:19 karolherbst: maybe it is somewhere in the vbios, maybe not
08:19 SaveTheRobots: seems like a relatively unimportant thing for nvidia to release openly :\
08:20 karolherbst: well
08:20 karolherbst: there might be IP problems
08:20 karolherbst: or internal problems
08:20 SaveTheRobots: yeah... good point
08:20 karolherbst: but the factors I REed so far, seems to work real nice for most gpus
08:20 karolherbst: but I fear there are chipset specific factors as well
08:21 SaveTheRobots: yeah, i remember trying nouveau a while back and it was basically unusable for gaming, it's come a LONG way thanks to awesome people like you
08:21 SaveTheRobots: i wish there was more i could do to help
08:21 SaveTheRobots: (besides testing)
08:22 karolherbst: mhh
08:22 karolherbst: we need a tool to get the blob driver to a specific clock
08:23 karolherbst: maybe
08:23 karolherbst: maybe not
08:23 karolherbst: ohh I know
08:24 karolherbst: you could mess with the vbios a bit and force nvidia to set the clock
08:24 karolherbst: but for that you need to install envytools
08:24 SaveTheRobots: already done :p
08:24 SaveTheRobots: i had to install it to get your tool working
08:24 pmoreau: karolherbst: LnkSta: Speed 8GT/s, Width x8 on 0f with the blob, for my Kepler card, which you asked for some time ago
08:24 karolherbst: pmoreau: ...
08:24 karolherbst: k
08:25 karolherbst: pmoreau: I already feared this might happen
08:25 pmoreau: As for the clocks: core: 135MHz-925MHz, memory: 5016MHz
08:25 karolherbst: seems like I have to take another look at how we know which pcie speed to set
08:25 SaveTheRobots: karolherbst: i gotta shoot out but if there's a guide or whatever that explains what i need to do to force clock speeds then let me know and i'll give it a try
08:25 karolherbst: SaveTheRobots: well there is a tool called nvafakebios, but it needs to be called before nvidia is loaded
08:26 karolherbst: SaveTheRobots: but we could do that later together when you are back
08:26 SaveTheRobots: i can do that, i don't use an initramfs and i boot for runlevel 3
08:26 SaveTheRobots: yeah sure :)
08:26 karolherbst: pmoreau: does the width changes with nvidia?
08:26 SaveTheRobots: i'll holla when i'm free
08:28 pmoreau: karolherbst: Nope, it always stays at a width of x8. Only the speed changes between 2.5 and 8.0
08:31 karolherbst: pmoreau: k
08:31 karolherbst: pmoreau: does it ever set it to 5.0?
08:31 karolherbst: like on the 0a pstate?
08:31 pmoreau: --" Alt-tabbing between applications makes the blob jump all the way from 07 to 0f… --"
08:32 pmoreau: Yes, 07: 2.5, 0a: 5.0, 0e and 0f: 8.0
08:32 karolherbst: k
08:32 karolherbst: thanks
08:32 karolherbst: pmoreau: this was the nve7?
08:32 pmoreau: Exactly
08:35 karolherbst: 19: 2.5, 28: 5.0, 00: 8.0. currently nouveau does & 0x3 the value and sets 0: 5.0, 1,3: 2.5, 2: 8.0 ....
08:35 karolherbst: well I guess I will play around this a little
08:37 pmoreau: K
09:34 mupuf: karolherbst: this is not really calibration. more like knowing what bandwidth you are interested in :)
09:35 karolherbst: mupuf: you mean the power sensor stuff?
09:35 mupuf: yep
09:35 karolherbst: yeah I know
09:36 karolherbst: that's why I try to use the term "configure" instead
09:36 karolherbst: but the ina3221 can be calibrated
09:36 karolherbst: and then we could read out the power consumption directly from the sensor
09:39 karolherbst: mupuf: regarding the missing bit we miss on nvf0/nvf1 to read out the clock. It behaves rather strangly. Up to a specific clock nouveau reads it out right, then while nvidia increases the clock, the read out value decreases
09:40 mupuf: karolherbst: right, this is something we need to investigate
09:42 karolherbst: well SaveTheRobots seems to want to help with nouveau and he has a nvf1, I think I might give it a shot, no idea what I am searching for though
09:43 karolherbst: mhh
09:43 karolherbst: the nv_trim_hwref.h is quite different though
09:44 karolherbst: https://gist.github.com/karolherbst/fb215c285ed1d1665cfe
09:45 karolherbst: my guess would be that 0x13700c and 0x137018 are somewhat important here :D
09:45 karolherbst: and then this dynramp thing too, but not for reading out the clock
09:47 karolherbst: yeah
09:47 karolherbst: there is some crazy stuff on teh gk20a too
09:48 karolherbst: the clocks are slid down/up there
10:08 karolherbst: mupuf: but SaveTheRobots card has really odd coefficients in the vbios: c0 -7934361 c1 28908 c2 -100767 c3 0 c4 0 c5 0
10:09 mupuf: what are those?
10:09 karolherbst: voltage stuff
10:10 karolherbst: and this row equals 1.24V...
10:10 karolherbst: just never thought they would put in so messy values
10:23 SaveTheRobots: karolherbst: hey, i'm back and some time to spare if you work on this some more
10:23 karolherbst: SaveTheRobots: k, we could fake your vbios then
10:26 SaveTheRobots: sure :]
10:26 SaveTheRobots: i have envytools installed
10:27 karolherbst: SaveTheRobots: before nvidia gets loaded: nvafakebios -e 6fed:2b -e 6fee:08 $path_fo_vbios
10:27 karolherbst: then just continue booting as normal
10:28 SaveTheRobots: k, 1 mo
10:34 SaveTheRobots: karolherbst: done
10:35 karolherbst: SaveTheRobots: open nvidia-settings and set the performance mode to max perf
10:35 karolherbst: it should then clock to 1046 MHz
10:35 SaveTheRobots: 980 Mhz
10:36 karolherbst: mhh are yuo sure nvidia wasn't loaded while executing nvafakebios?
10:36 karolherbst: ohhh wait
10:36 karolherbst: I gave you the wrong arguments
10:37 SaveTheRobots: ah :p
10:37 karolherbst: nvafakebios -e 6ff0:2b -e 6ff1:08 $path_fo_vbios
10:37 SaveTheRobots: do i need to reboot or just unload the modules?
10:37 karolherbst: unloading should be fine
10:37 SaveTheRobots: k, 1 mo
10:40 SaveTheRobots: done, i'm at 1045 Mhz
10:41 karolherbst: nice
10:41 karolherbst: then run my tool again
10:42 SaveTheRobots: done
10:42 SaveTheRobots: current voltage (??V), expected voltage (??V), abs diff (??V),rel diff nouveau/nvidia (%), pstate, cstate, temperature(??C)
10:42 SaveTheRobots: 1062500, 1066713, 4213, 100.396518, 15, 32, 58
10:42 SaveTheRobots: 1062500, 1066713, 4213, 100.396518, 15, 32, 57
10:42 SaveTheRobots: last 2 lines repeat
10:42 karolherbst: mhhh
10:43 karolherbst: I am running out of ideas now
10:43 karolherbst: seems like the voltage is fine
10:43 SaveTheRobots: http://dpaste.com/2BQ4AP0
10:44 SaveTheRobots: do i need to run a game to load the gpu ?
10:44 karolherbst: no
10:44 SaveTheRobots: k
10:44 karolherbst: but why did nouveau volt to 1.05V
10:44 karolherbst: I really doubt that such a small difference makes the gpu that unstable
10:44 karolherbst: but we shouldn't volt so low anyway
10:45 SaveTheRobots: i've tried boost 1 3-4 times now, 3 times it crashed, the 4th time, it was stable
10:45 SaveTheRobots: (while gaming)
10:45 karolherbst: mhh
10:46 karolherbst: I know where 1.05V comes from
10:46 karolherbst: at 70°C the voltage drops that low
10:46 karolherbst: SaveTheRobots: still on nvidia?
10:46 SaveTheRobots: yup
10:46 karolherbst: nvaforcetemp 70
10:46 karolherbst: and then run the tool
10:47 SaveTheRobots: 1050000, 1086831, 36831, 103.507714, 15, 34, 61
10:47 SaveTheRobots: 1050000, 1086831, 36831, 103.507714, 15, 34, 70
10:47 SaveTheRobots: (left the tool running)
10:47 karolherbst: mhh did the clock change?
10:47 SaveTheRobots: nope, 1045Mhz
10:48 karolherbst: nvaforcetemp 20
10:48 SaveTheRobots: 1062500, 1066713, 4213, 100.396518, 15, 32, 20
10:48 SaveTheRobots: 1045 Mhz
10:48 karolherbst: okay....
10:48 karolherbst: mupuf: did you follow?
10:49 karolherbst: it seems like nvidia reclocks the cores on higher temps
10:49 karolherbst: like in, it configures the set clock different for the same clock
10:50 karolherbst: SaveTheRobots: nvapeek 0x13700c; nvapeek 0x137018
10:50 enaken: can't write file sudo echo 20 > /sys/kernel/debug/dri/0/pstate
10:51 karolherbst: SaveTheRobots: then nvaforcetemp 70; nvapeek 0x13700c; nvapeek 0x137018
10:51 karolherbst: enaken: as root
10:51 enaken: sudo
10:51 karolherbst: enaken: well echo 20 is run as root
10:51 karolherbst: but a user can't write into debugfs
10:51 karolherbst: the output is written as non root with sudo
10:51 SaveTheRobots: 0013700c: 02000000
10:52 SaveTheRobots: 00137018: 00060000
10:52 SaveTheRobots: (20C)
10:52 karolherbst: SaveTheRobots: mhh
10:52 SaveTheRobots: 0013700c: 02000000
10:52 SaveTheRobots: 00137018: 00060000
10:52 SaveTheRobots: 70C
10:52 karolherbst: SaveTheRobots: you have to wait a little with 70
10:52 SaveTheRobots: k
10:52 karolherbst: 1 second is enough though
10:52 karolherbst: nvidia just needs time to reclock
10:52 karolherbst: or to detect the temperature difference
10:53 SaveTheRobots: 0013700c: 02000000
10:53 SaveTheRobots: 00137018: 00060000
10:53 karolherbst: damn...
10:53 karolherbst: k, then brute force it is
10:54 karolherbst: SaveTheRobots: go to 20°C
10:54 karolherbst: then nvapeek 0x137000 0x1000
10:54 karolherbst: go to 70°C
10:54 karolherbst: nvapeek 0x137000 0x1000 again
10:54 karolherbst: paste both outputs on patebin or somewhere
10:54 karolherbst: enaken: I hope you understood what I meant?
10:55 enaken: karolherbst: sudo -s
10:55 karolherbst: yeah, one possibility
10:55 enaken: karolherbst: and echo 20 > /sys/kernel/debug/dri/0/pstate
10:55 enaken: karolherbst: but "write error"
10:56 karolherbst: did you check if your gpu supports the 20 pstate?
10:56 karolherbst: ohh wait
10:56 karolherbst: does the file /sys/kernel/debug/dri/0/pstate actually exist?
10:56 SaveTheRobots: karolherbst: that nvapeek caused my display to go white and crashed my machine
10:56 SaveTheRobots: cant ssh in either
10:56 karolherbst: SaveTheRobots: ...
10:57 enaken: karolherbst: cat /sys/kernel/debug/dri/0/pstate
10:57 enaken: exist
10:57 karolherbst: SaveTheRobots: it shouldn't though, I doubt that reading any of this causes lockups...
10:57 SaveTheRobots: shall i try again?
10:57 karolherbst: yeah
10:57 karolherbst: SaveTheRobots: maybe save it to a file
10:58 karolherbst: it could be that the machine still saves it
10:58 enaken: karolherbst: http://paste.ubuntu.com/15511790/
10:58 enaken:
10:58 karolherbst: enaken: do you only get a "write error"or something else?
10:59 enaken: karolherbst: echo: ошибка записи: Функция не реализована
10:59 karolherbst: k. -ENOSYS
10:59 karolherbst: enaken: what gpu do you have?
11:00 enaken: karolherbst: 8600M GS
11:00 karolherbst: G84
11:00 karolherbst: mhh
11:00 karolherbst: no idea if those support reclocking
11:00 karolherbst: but I think not
11:01 enaken: karolherbst: yes, https://nouveau.freedesktop.org/wiki/PowerManagement/
11:01 karolherbst: and where does it state, that G84 supports reclocking?
11:02 karolherbst: I am sure tesla reclocking on works for G94+ chips
11:03 enaken: karolherbst: mostly.. not all)
11:03 SaveTheRobots: karolherbst: 20C http://dpaste.com/1MCGC3R
11:03 SaveTheRobots: try to forcetemp again reports: "chipset unsupported"
11:03 SaveTheRobots: guess i need to reboot again
11:04 karolherbst: SaveTheRobots: do you do this while the gpu is at 1046MHz?
11:04 karolherbst: *1045
11:04 SaveTheRobots: yup
11:05 karolherbst: k
11:10 SaveTheRobots: karolherbst: 70C http://dpaste.com/27SJQJE
11:10 karolherbst: SaveTheRobots: awesome, thanks
11:10 karolherbst: ahhh
11:10 SaveTheRobots: np
11:10 karolherbst: yeah, that looks usefull
11:11 SaveTheRobots: sweet :]
11:11 SaveTheRobots: need anything else?
11:13 karolherbst: not yet
11:13 SaveTheRobots: k
11:13 enaken: karolherbst: do you know another way to downclock G84?
11:13 karolherbst: not really
11:22 SaveTheRobots: karolherbst: i ran the 20C nvapeek again out of curiosity and it changed VERY slightly: http://dpaste.com/1WN03TV
11:23 SaveTheRobots: not sure if that means anything, it's pretty much just line 25 i think
11:23 karolherbst: yeah, but it is the same as the 70 one
11:24 karolherbst: except the first two lines
11:24 SaveTheRobots: ahh ok
11:24 karolherbst: 0013701c seems to be important
11:25 SaveTheRobots: and line 25
11:25 SaveTheRobots: but maybe that's just noise
12:01 Tom^: SaveTheRobots: well, i never used the dynamic volts on 0f either, but ask karolherbst and see if he has a clue.
12:01 Tom^: SaveTheRobots: i had a silly workaround that always set the volt to max available , 1.2v :p
12:31 TECTEP: Hello everybody, I have a problem with Display connected to DVI output of docking station DELL. When laptop lid is closed, BIOS starts the DVI output, I see bios messages etc. When noveau/lightdm start, DVI output is stopped(dsplay goes in standby). xrandr --prop reports EDID correctly, I enable both ext display and LVDS display in 'arandr' ... no DVI signal. Any debug ideas ?
12:32 TECTEP: No error messages in kernel log, or Xorg.0.log
12:32 seiyh: no ideas
12:32 TECTEP: :-(
12:32 seiyh: not a programer
12:32 seiyh: use oss4?
12:38 rester: TECTEP: use oss4?
12:38 TECTEP: Observed also on other DELL laptops with Nvidia chipset, not combination intel/nvidia
12:43 pmoreau: TECTEP: Which NVIDIA chipset do you have? You could boot with `drm.debug=0xf` on the kernel command line, which should information about screen hotplugging & co IIRC. Maybe add as well `nouveau.debug=debug` on the kernel command line.
12:44 TECTEP: just a sec
12:44 pmoreau: And giving a link to your Xorg.log and dmesg would still be interesting :-)
12:44 TECTEP: 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation G84GLM [Quadro FX 1600M] [10de:040d] (rev a1)
12:44 TECTEP: Subsystem: Dell Device [1028:019b]
12:44 TECTEP: Kernel driver in use: nouveau
12:45 TECTEP: Linux tester-dell 3.13.0-39-generic #66-Ubuntu SMP Tue Oct 28 13:30:27 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux
12:45 TECTEP: cat /etc/issue
12:45 TECTEP: Ubuntu 14.04.3 LTS \n \l
12:47 pmoreau: Oh, 3.13…
12:47 karolherbst: TECTEP: install the xenial LTS kernel
12:47 TECTEP: PITA with 3.13 ?
12:47 TECTEP: ;-)
12:48 pmoreau: Nah, more like: "it's super old!!"
12:49 pmoreau: And, no one wants to debug an issue on such an old kernel, since it could have been already fixed in a newer release
12:49 TECTEP: could be. But vanilla kernel (cloned from Linus's tree) results in the same... The same observerd also with Debian testing installed and live USB
12:50 pmoreau: However, there seem to be issues with external screens on Tesla, not just G84. I have some with my MCP79 (NVAC), and IIRC, Hans had some on G96.
12:50 TECTEP: what is xenial LTS ? can I run it from usb (live) ?
12:50 karolherbst: TECTEP: install linux-generic-lts-xenial or linux-generic-lts-wily
12:50 karolherbst: xenial is 4.4, willy is 4.2
12:50 TECTEP: ok, I'll google it
12:51 pmoreau: Well, if you tried a clone from Linus' tree recently, no point in trying a recent LTS I guess
12:51 karolherbst: yeah well
12:51 TECTEP: last test was ~ 3 weeks ago ( git pull --rebase, make old config, make deb ...)
12:51 karolherbst: depends on the configuration
12:51 karolherbst: mhh
12:51 TECTEP: :-)
12:52 karolherbst: no idea if oldconfig pulls the current kernel config
12:52 TECTEP: it asks for new items
12:52 karolherbst: well you get less pain with the lts kernels from your package manager anyway
12:53 pmoreau: Otherwise, you can try images from https://nouveau.pmoreau.org/ if you don't want to update your laptop. Sadly, I don't have the very latest ones…
12:53 karolherbst: TECTEP: did you check you are on 4.5 after installing from git?
12:53 TECTEP: no problem
12:53 TECTEP: I'll see to download something
12:56 TECTEP: now I run from live USB stick ..., mounted ... I see last try was with kernel 4.3.0-1-amd64
12:59 TECTEP: @ karolherbst, pmoreau: Thanks guys, I'll download some stuff from https://nouveau.pmoreau.org/ and try it :)
12:59 pmoreau: If you could paste your dmesg and Xorg.log with the previously mentioned command line arguments, that could maybe help
12:59 pmoreau: But, we'll probably need an MMIOtrace of NVIDIA's driver.
13:00 pmoreau: I need to have a look at a similar issue on my laptop at some point.
13:04 TECTEP: Added drm.debug=0xf nouveau.debug=debug debug initcall_debug to kernel line, rebooting now ...
13:07 karolherbst: mupuf: when you got time, could you plug a maxwell to reator?
14:37 imirkin: anyone with a maxwell gpu, could i get the output of 'glxinfo -l -s' against mesa 11.2-rc3?
15:17 karolherbst: mupuf:
15:17 karolherbst: ...
15:17 karolherbst: I am pretty sure it is the 1c reg
15:18 karolherbst: 137004: 00012a08 0013701c: 41002216 == 137004: 00012b08 0013701c: 41002316 (equal clock set)
15:42 dcomp_: trying to decode the i2c bitbang in an mmiotrace ... not sure what I'm looking for
16:02 RSpliet: karolherbst: it seems my NVCE card has a BOOST table version 0x10, nvbios refuses to parse it
16:02 RSpliet: mind looking into that in due time?
16:04 karolherbst: RSpliet: yeah
16:04 karolherbst: is it uploaded already?
16:04 RSpliet: yep
16:04 karolherbst: ahh found it I guess
16:04 karolherbst: k
16:04 RSpliet: I hope it's trivial, but you might have all this fresher in memory than I do :-)
16:14 karolherbst: mhh
16:14 karolherbst: header length 4
16:15 karolherbst: mhh
16:15 karolherbst: I would have to guess
16:18 karolherbst: RSpliet: well there are also no cstates
16:18 karolherbst: and the boost table has no entries
16:18 karolherbst: boosting wasn't a thing on Fermi anyway, though some fermi cards have some tables for it
16:19 karolherbst: this fermi also has only 4 voltage map entries
16:19 karolherbst: one for each pstate and a 0 one
16:19 karolherbst: so even if there was something in the boost table, it would be useless
16:20 karolherbst: RSpliet: but you could RE the unknowns: -- ID = 2: voltage_min = 962500, voltage_max = 1050000 [µV] unks: 28659874 -20706 55852--
16:20 karolherbst: if you want
16:20 karolherbst: there are coefficients like the one from kepler: c0 * f0 + c1* f1 + c2 * f2
16:20 karolherbst: c0-c2 are those values
16:26 RSpliet: in due time
16:28 RSpliet: heh, that boost table is more empty that I anticipated :-P
16:29 karolherbst: yeah
16:29 karolherbst: and doesn't make sense
16:29 karolherbst: 0 sub count
16:29 karolherbst: 10 sub length
16:29 karolherbst: or was it the other way around?
16:29 karolherbst: and like 6 entries?
16:30 RSpliet: 0x10 version, 0x4 header length, 0x6 either entry length or count, the other is 0x0
16:30 karolherbst: I get the feeling that since kepler the vbios guys were less sloppy
17:03 karolherbst: uhhhh
17:03 karolherbst: pmoreau: I think I got it
17:04 karolherbst: maybe you see it too :D 0: 8.0 1: 2.5 2: 8.0 3: 2.5 4: 8.0 5: 2.5 ....
17:04 karolherbst: maybe it is a bit saying "max pcie" or "min pcie"
17:04 karolherbst: but then again
17:04 karolherbst: why does 0a say 5.0 for me
17:05 karolherbst: either way, the parsing is completly wrong
17:06 karolherbst: 38 gives me 5.0, but 18 8.0 allthough envytools does a & 0x03 mask
17:10 karolherbst: ohhh wait I have a stupid idea
17:15 karolherbst: what the...
17:15 karolherbst: field & 0x01 => 2.5
17:15 karolherbst: field & 0x20 => 5.0
17:15 karolherbst: else => 8.0
17:15 karolherbst: lowest wins
17:16 karolherbst: if somebody has a better idea please say it: https://gist.github.com/karolherbst/1c1dfb1ea50ad2a17e76
17:19 karolherbst: I am sure though there might be a bit selected 8.0
17:19 karolherbst: and default means max or something
18:17 imirkin: omg... i'm going to murder someone
18:17 imirkin: finally figured out the reason for that WoW thing
18:24 RSpliet: imirkin: please tell me it's not related to MUL+ADD->MAD folding?
19:31 imirkin: RSpliet: hahahahah
19:31 imirkin: RSpliet: no, it's skeggsb's fault
19:31 imirkin: (from a very long time ago)