02:39hakzsam: imirkin, okay, makes more sense
06:20pmoreau: karolherbst: Still stable on the pmoreau branch; I get a score of 68 in Unigine Heaven… --"
06:20karolherbst: uhh nice
06:20karolherbst: pmoreau: but the clock is higher now, right?
06:21pmoreau: It is
06:21pmoreau: Oh, you mean with boost?
06:21pmoreau: Haven't tried boost
06:21karolherbst: the boost stuff isn't on your branch
06:21pmoreau: Ah, k
06:21karolherbst: besides, it only lowers clocks
06:22pmoreau: Well, I still have the 925MHz core clock, like I did yesterday IIRC
06:22karolherbst: ohh
06:22karolherbst: sure you switched the branch?
06:22pmoreau: Let me check
06:22karolherbst: dmesg | grep nouveau | grep MHz
06:23karolherbst: if it shows boost/base clocks, you are still on the kepler branch
06:23karolherbst: we should move some stuff out of init into oneinit ...
06:23pmoreau: The grep returns nothing
06:23karolherbst: resuming the gpu prints some stuff we don't need
06:23karolherbst: pmoreau: mhh then let me check
06:24karolherbst: pmoreau: cat pstate
06:25karolherbst: the boost table says your base/boost clock is 925MHz, which is weird, because they are the same :D
06:25karolherbst: but that has no effect whatsoever
06:25pmoreau: 07: core 270-405 MHz memory 838 MHz AC DC * 0a: core 270-925 MHz memory 1560 MHz
06:25pmoreau: 0e: core 270-925 MHz memory 4000 MHz 0f: core 270-925 MHz memory 5016 MHz
06:25pmoreau: DC: core 405 MHz memory 837 MHz
06:25karolherbst: why are the pstates limited...
06:26karolherbst: ohh i know
06:26karolherbst: stupid boost tables
06:26karolherbst: this vbios is just a mess
06:26karolherbst: your gpu has not GPUBoost :p
06:26karolherbst: it behaves completly like a tesla with cstates :D
06:27pmoreau: lol
06:27karolherbst: even the normal fermi has more boost stuff
06:27karolherbst: they really ruined this gpu
06:28karolherbst: pmoreau: can you run gputest furmark and see how hot your gpu gets?
06:28pmoreau: (BTW, do you know which identifier I need to use for the Intel driver in the xorg.conf? I'd like to set it to use DRI3)
06:28pmoreau: I'll check that
06:29karolherbst: pmoreau: use mine xorg.conf https://gist.github.com/karolherbst/1f1bdd1a3822df74097f
06:29karolherbst: this way you can unload nouveau while X is running
06:29karolherbst: you only need the dummy driver
06:30pmoreau: Thanks
06:30karolherbst: though it might that you also need a X patch
06:31karolherbst: ohh no, it was merged in 1.18
06:31pmoreau: I'm on 1.18.x
06:31pmoreau: Should be good then
06:31karolherbst: yeah
06:31karolherbst: with 1.17 loading/unloading nouveau hanged my X server
06:31karolherbst: if nouveau was added lateR? no idea
06:31karolherbst: but there was an issue :)
06:35pmoreau: 85C it seems
06:37karolherbst: mhh
06:37karolherbst: k, this is hotter than mine
06:38karolherbst: so we shouldn't play around witht he clocks then :D
06:38karolherbst: as I said, like a tesla :D
06:39pmoreau: But high is 95C and crit is 105C, so still a bit of room there
06:39karolherbst: pmoreau: yeah, but you only do rendering now
06:39karolherbst: imagine you also do video decoding and encoding
06:40karolherbst: pmoreau: anyway, there is one point left which bothers me a bit. the vbios declares 5.0 PCIe speed on the higher pstates
06:40karolherbst: it would be interessting to know what nvidia does
06:40karolherbst: and if nvidia changes the width
06:41karolherbst: LnkSta is the important line in lspci
06:41karolherbst: if nvidia changes the width, then it would be nice if you would do a mmiotrace of that
06:42karolherbst: just so that we have at least one for kepler
06:42karolherbst: and if it changes the link to 8.0 we might read it out wrongly out of the vbios
06:44karolherbst: pmoreau: I am curious why you get only 68 points though
06:44karolherbst: with mine I get like 340
06:45karolherbst: pmoreau: which resolution?
06:45Tom^: does he still have that .drirc crap that got merged a few months ago
06:45karolherbst: I doubt it
06:45Tom^: because well, once the .drirc has it added in ~ it wont be removed unless you do it :p
06:45karolherbst: most likly he runs it at an insane resolution
06:46karolherbst: Tom^: why should that be?
06:46karolherbst: usually you don't have any .drirc
06:46karolherbst: only the system-wide one
06:46Tom^: no idea why i had one
06:47Tom^: perhaps driconf or something like that creates one
06:47karolherbst: yeah
06:48karolherbst: pmoreau: your gpu is just too slow... 750M was it?
06:49karolherbst: mhh
06:49karolherbst: it should be clocked at 967MHz boost I think
06:50karolherbst: yeah well, most likely the apple guys just remove some high perf stuff, so that the gpu stays a bit cooler
06:57pmoreau: karolherbst: At 2880x1800
06:57karolherbst: pmoreau: ohh k
06:57karolherbst: then the perf difference is to be expected
06:58pmoreau: Finally on the Intel chipset! Thanks for your Xorg.conf
06:58karolherbst: no problems :)
07:00pmoreau: Yeah! And with DRI3, so I can switch between both!
07:00pmoreau: Well, not really switch, but well
07:00karolherbst: yeah I know :p
07:00karolherbst: but
07:00karolherbst: and this is a big bug
07:00karolherbst: 5.0 pcie is a bit slow..
07:00karolherbst: pmoreau: try fullscreen glxgears or something
07:00karolherbst: you will be disappointed
07:01karolherbst: prime offloaded I mean
07:02pmoreau: 2866FPS -> 58FPS
07:02karolherbst: pmoreau: I get around 520 fps with 8.0 at fullhd
07:04pmoreau: And vulkaninfo correctly founds the Intel chipset!
07:05pmoreau: Time to run some VUlkan demo then
07:06karolherbst: pmoreau: ohh seems like 5.0 doesn't make much of a difference here
07:06karolherbst: 344/540-550/550-590 on 2.5/5.0/8.0
07:26karolherbst: pmoreau: ohh and it would be also good to know what nvidia clocks to at full load
07:29pmoreau: Ok, I put it on my todo list
07:34karolherbst: thanks a lot
08:18karolherbst: pmoreau: do you have anything against if I put tested-by tags ont he patches?
08:18pmoreau: Not at all, please do
08:19karolherbst: and I really don't like journald if you need to debug errors...
08:19karolherbst: the only way to get stuff is pstate
08:19karolherbst: ..
08:19karolherbst: pstore
09:05imirkin: pmoreau: karolherbst: do either of you have a minute to trace a couple of dEQP tests?
09:05karolherbst: I would have to setup dEQP first
09:07imirkin: oh, i thought you had that done already.
09:08karolherbst: do I have to do something special or just grab the deqp sources and build it?
09:10karolherbst: imirkin: which DEQP_TARGET should I set?
09:10karolherbst: x11_glx or x11_egl_glx?
09:17imirkin: karolherbst: either
09:17imirkin: but not both :)
09:17imirkin: it'll fail with some idiot libpng thing
09:18imirkin: so you need this patch: http://hastebin.com/ehetupikix.avrasm
09:18imirkin: also, if you don't have xorg 1.18.1 you need another patch, but i suspect you're on the latest and greatest
09:19karolherbst: I am on 1.18.1
09:19imirkin: ok you're good on that front then
09:19karolherbst: there is some really annoying freeze with 1.18.2 and chromium
09:20karolherbst: yeah, already stumbled across the libpng thing
09:20karolherbst: this is super annoying, they could just use the system one
09:20imirkin: i sent a patch
09:20imirkin: they've allegedly integrated it
09:20imirkin: which means it should be publicly visible in the next 5 years or so.
09:20karolherbst: ohh
09:21karolherbst: where is the patch? :D
09:21imirkin: see above :p
09:21karolherbst: ahhhh
09:21karolherbst: right
09:21karolherbst: my fault
09:22imirkin: you're grabbing from https://android.googlesource.com/platform/external/deqp/ right?
09:22karolherbst: yeah
09:26imirkin: ok that's right then
09:27pmoreau: imirkin: Not right now sorry.
10:08karolherbst: imirkin: I get this error: "print "Both --name and --id must be specified""
10:08karolherbst: ohhh
10:08karolherbst: python2....
10:10mupuf: karolherbst: hey, could you give me the output dmesg when running your kernel?
10:10karolherbst: mupuf: all?
10:10mupuf: just the nouveau part
10:10mupuf:is reviewing your patch series
10:11mupuf: sunday is patch review day, isn't it?
10:11karolherbst: mupuf: https://gist.github.com/karolherbst/50834676785ad7206f70
10:11karolherbst: ahh nice
10:12karolherbst: mupuf: pmoreau only needs 5 patches of those: https://github.com/karolherbst/nouveau/commits/pmoreau
10:12karolherbst: mupuf: and I was thinking we should at least get those merged for the next release, because they are trivial and already help somebody
10:12mupuf: yeah, not a big fan of how you print the boost and base clocks
10:12mupuf: well, talk to ben :D
10:13karolherbst: mupuf: his vbios doesn't contain the coefficients stuff, and info.min == info.max, so there is nothing we can do wrong
10:13mupuf: My work is done right? I already reviewed these patches
10:13mupuf: but seriously, don't try to hurry it
10:13karolherbst: only some of them
10:13karolherbst: these are all now
10:13mupuf: pmoreau is a grown up
10:13mupuf: he can recompile his kernel :p
10:13karolherbst: :D
10:22Yoshimo: so karol, if the 4.6 merge window is over, will there be any patches on your github left that aren't upstream yet?
10:34karolherbst: Yoshimo: many
10:34karolherbst: like most of them
11:15jayhost: karolherbst you still want me to keep gm107 on high clocks?
11:16karolherbst: jayhost: if nothing happened until now, then it is all fine
11:16karolherbst: you can stop now :D
11:27jayhost: Alright.
12:30karolherbst: mupuf: should I have one debugfs file for each clock? base_mhz, boost_mhz? And should I put in there the real or the doubled clock (or is it real and halfed?)? I would got for each clock one file, because everybody else seems to do that
12:35mupuf: hmm, you are confusing me
12:35karolherbst: which part is confusing?
12:37mupuf: sorry, back
12:38mupuf: why would you want to have as many files as clock?
12:38mupuf: I would propose pstate to actually state the boost level
12:38mupuf: like
12:39mupuf: boost level: 0 (1123 MHz), *1 (1175 MHz), 2 (1200 MHz)
12:40karolherbst: mhhh
12:40mupuf: and create a debugfs file to change the boost level you want also
12:40mupuf: or ... just create this boost level file
12:40mupuf: and make it like pstate
12:40karolherbst: well there is no difference between selecting a cstate
12:40mupuf: makes sense?
12:41karolherbst: boosting on nvidia gpus is only a thing, because we have a base clock entry inside the vbios, without it, there would be no such thing as "boosting", because in the end it is just reclocking depending on the load
12:42karolherbst: (+ some volt/temperature stuff, but that doesn't make much of a difference)
12:42karolherbst: in the end at any given time we have a list of cstates we can reclock to
12:43mupuf: yeah
12:43karolherbst: and this list may have more or less entries depending on the current temperature
12:43mupuf: so, what about: max_cstate ?
12:43karolherbst: well I could also just revice my cstate debugfs file
12:43karolherbst: and we just push out a list with all cstates
12:43mupuf: not really no, please
12:43mupuf: I just want to expose your NvBoost parameter
12:44karolherbst: k
12:44karolherbst: what we could do though is this
12:44mupuf: does that make sense to you?
12:44karolherbst: debugfs file name "boost" or "gpu_boost"
12:44karolherbst: 0: base MHz
12:44karolherbst: 1*: boost MHz
12:44karolherbst: 2: max MHz
12:44karolherbst: and we just put the value in there
12:46karolherbst: on mine it would be 0: 705 MHz 1: 797 MHz 2: 862 MHz
12:46karolherbst: where the max one might not be even reached, because of the voltage calculations
12:47karolherbst: ohhh
12:47karolherbst: wait
12:47karolherbst: I think this was what you meant :D
12:55mupuf: * 1: boost MHz
12:55mupuf: yes
12:55mupuf: this is what I meant by : max_cstate
12:55mupuf: not a guarantee, just a limiter :)
12:56karolherbst: yeah I think it would make actually sense to have
12:56karolherbst: *this
12:56mupuf: yep :)
12:57mupuf: The fact that you wanted to expose it in the kernel logs pretty much said it already
12:58karolherbst: well there are cards where this file would be of noe use though :D
13:02mupuf: no worries
13:02mupuf: at least, no need to reboot to test the boost clocks, that's a win!
13:03karolherbst: I think we need a field in nvkm_clock to remember the cstate we _want_ to have, because the one we actually have might be lower, and in our volt adjustmen code we could then clock up if possible
13:03mupuf: oh, right
13:04pmoreau: God it was complicated to install Linux on the computer: first manage to connect a screen to one of the GPU, then the ISO had old version of the GPG keys and didn't wanted to install the packages, once I had the new version, internet didn't wanted to work… --"
13:04karolherbst: but currently there are all kind of states saved there
13:04karolherbst: and I really want to clean up that mess
13:04karolherbst: astate, tstate, dstate?
13:04karolherbst: ...
13:04karolherbst: and then ustate_ac and ustate_dc
13:05karolherbst: pmoreau: you should try to install gentoo for once :D
13:05karolherbst: or better: LFS
13:06pmoreau: :-D
13:06pmoreau: It's more a "it was complicated" because everything was working against me, not the task per se was complicated.
13:06karolherbst: but the gentoo bootstrap isos are crap and garbage
13:06karolherbst: they don't have any efi stuff on it
13:07karolherbst: uhh
13:07karolherbst: pmoreau: never forget, I installed gentoo once on my mbp, from scratch
13:07karolherbst: it was not a retina one, but basically the same issues
13:07pmoreau: I'm considering installing gentoo one day
13:07karolherbst: bios emulation: only radeon card, configureing efi crap, moving to intel as main ;) much fun
13:08pmoreau: :-)
13:08karolherbst: and this was before vga_switcheroo supported the gmux
13:09karolherbst: pmoreau: I suggest you to use a live CD that you trust for this :D
13:09pmoreau: Trying to install Fedora/Debian/Ubuntu in nconfig on my old mbp was kinda fun as well
13:09pmoreau: s/nconfig/ncurvers
13:09pmoreau: ncurves
13:09karolherbst: at some point I shall try out these non APT based distributions... never did that actually
13:13mupuf: pmoreau: ncurses?
13:13mupuf: funny. arch decided to do away with any kind of GUI to install
13:14mupuf: to be fair, it is not like it is needed
13:15pmoreau: mupuf: Yes, ncurses, thanks
13:17Tom^: mupuf: it was just a blue menu driven ncurses menu that started fdisk and nano for you. it was a bit pointless imo :P
13:17mupuf: Tom^: agreed
13:46kugel: karolherbst: pong
13:46kugel: sorry, very busy currently
13:47karolherbst: kugel: ahh, k, just wanted to know if you still have any reclocking issues with nouveau
18:01karolherbst: mupuf: still there?
21:50t4nk270: someone else is having too much lag even with the mouse using gtx960 with wayland?
21:51t4nk270: i think there isn't hardware accel.
22:05imirkin: t4nk270: that's right - there isn't. GM20x requires signed firmware in order to have acceleration, and nvidia has only very recently released proper information about it
22:06imirkin: t4nk270: 4.6-rc1 will have some acceleration support
22:08t4nk270: great news
22:10t4nk270: thank you :)