01:41karolherbst: we really should fix those fuc sources... or at least improve their code
01:45huelter: those fucing sources right
01:51imirkin: fuc = falcon microcode
01:52karolherbst: he knows, he just tried to be funny .D
01:57imirkin: ah. probably too tired to tell.
02:26huelter: imirkin, I think karolherbst work solved the issue here
02:26huelter: no hangs so far
02:27huelter: his v4 branch that is under revision
04:55imirkin: skeggsb: pretty sure people have been using GM108 just fine by just treating it as GM107...
04:57skeggsb: yeah well, whichever trace i'm looking at - there's things to fix :)
04:58imirkin: i looked over a lot of the goldens, they all looked the same
04:58imirkin: but i also skipped a bunch
04:58imirkin: coz i didn't really know what was what :)
04:58skeggsb: yeah, that stuff all looked ok when i compared traces
07:46huelter: imirkin: Well, talked too soon... even reclocking did not stop the hangs. Also managed to get a kernel log, and submitted a bug report https://bugs.freedesktop.org/show_bug.cgi?id=95031
12:43hondz: Hi ! Videos play fine with xorg but I am trying to play them using fbdev and mpv/mplayer
12:43hondz: And it says it can't initialize video driver
12:43hondz: Arch is my distro
12:44karolherbst: I thought like fbdev is dead?
12:44hondz: karolherbst: I want to play videos without starting xorg...
12:45pq: hondz, you also do not want any kind of hardware acceleration, and enjoy tearing?
12:45karolherbst: I don't think many care about fbdev much anymore
12:45karolherbst: and wayland is a proper replacement for it
12:45karolherbst: at least in my eyes
12:46karolherbst: but if you really want to use fbdev, we could try to figure out your problem. I would just suggest to you to give wayland a try before
12:46hondz: karolherbst: let me see wayland
12:46karolherbst: well usually you can just install weston and start it from a tty
12:46karolherbst: weston is a compository for wayland
12:48hondz: karolherbst: it says weston-launch must be run from a virtual terminal
12:48pq: hondz, just curious, why is "without starting xorg" a worthy goal on its own right for you?
12:48karolherbst: yeah, run weston-launch then
12:48hondz: pq: There may be a solution to my problem
12:49hondz: pq: let me tell you my problem
12:49karolherbst: I think "weston" should be used in X and "weston-launch" from a tty
12:49hondz: pq: I am currently using arch without any DE or DM or whatever
12:49hondz: pq: And I like it to be terminal only
12:49hondz: pq: If I launch X, I hate how it looks
12:49karolherbst: mhh yeah, wayland sounds like the best idea in that case
12:50hondz: karolherbst: Hold on
12:50hondz: I launched x and I can't type anything...
12:50pq: hondz, I still don't understand what bad did xorg do to you, considering it is not responsible at all of how things look like.
12:50karolherbst: yeah, you need to use startx
12:51hondz: karolherbst: that's what I did
12:51karolherbst: startx setups twm/xterm/xclock for you
12:51karolherbst: but yeah, plain X is gruel
12:51hondz: wayland looks like arch looks without a dm ?
12:51pq: wayland also does not have a look
12:52karolherbst: right, but I guess most distros add some branding to weston
12:52pq: DEs have a look, whether they run using Wayland or X11
12:52hondz: I don't awnt a De...
12:52hondz: DE *
12:52karolherbst: well weston is only a compository
12:52karolherbst: which manages the wayland screen if you so will
12:52pq: weston is also a DE. A very crude and simple, but it is.
12:53karolherbst: uhh I wouldn't be so sure about that
12:53karolherbst: a DE should at least come with a power manager and a session environment
12:53karolherbst: weston completly lacks this
12:53hondz: I don't like it guys
12:53hondz: I want to use fbdev
12:54hondz: This may sound weird but I kinda need something specific
12:54pq: hondz, you do realize that fbdev is the only from fbdev,wayland,x11 that forbids hw accel, and has no way to prevent tearing, right? These are not an issue for you?
12:54pq: *only one
12:55karolherbst: at least in wayland you can run X clients :D
12:55hondz: pq: that's ok for me
12:55karolherbst: and fbdev has to be supported by the applications
12:56pq: hondz, ok. You one of the very very few people who feel like that and want fbdev.
12:56karolherbst: well maybe something is odd with nouveau
12:56hondz: pq: :) I know Richard Stallman irl if that puts the puzzle together
12:56karolherbst: hondz: could you provide us with your full dmesg output?
12:56pq: I don't really understand what "driver" the video player might be initializing for fbdev
12:56hondz: karolherbst: Going to take a picture and show you
12:56karolherbst: pq: a fb device
12:56karolherbst: I meant the full dmesg
12:57karolherbst: not only a few lines
12:57huelter: karolherbst: aboout yesterday https://bugs.freedesktop.org/show_bug.cgi?id=95031, managed to freeze and get a readable kernel log
12:58karolherbst: huelter: I guess there was nothing before that?
12:58huelter: before what?
12:59huelter: oh, the log
12:59karolherbst: huelter: you could boot with nouveau.config=NvVoltOffsetmV=100 for good messure
12:59karolherbst: and see if that improves the situation
12:59mwk: hmm, shouldn't it be possible to do it "properly" with egl?
12:59huelter: karolherbst: will do
12:59karolherbst: mwk: what do you mean?
12:59mwk: no bitstream decoding accel for you, but at least YUV -> screen should be accelerated
12:59mwk: karolherbst: no-DE video playback
13:00karolherbst: well, if he wants to use fbdev
13:00mwk: yeah, he wants to use fbdev, that's fine; I'm just wondering about no-DE playback in general
13:00pq: mwk, EGL does not run on fbdev anyway, contrary to what proprietary vendor attempt to suggest
13:00hondz: karolherbst: uploading the picture ,one moment
13:00karolherbst: yeah :/
13:01karolherbst: fbdev is a bit hard to handle
13:01karolherbst: becuase you have like, nothing?
13:01pq: or maybe someone hooked up software renderers, but meh.
13:01pq: mwk, I think you meant KMS
13:02hondz: karolherbst: http://imgur.com/v2rUV1L
13:02pq: writing a straight-to-KMS player would be well possible, and could use the YUV display if present.
13:02huelter: doesn't retroarch has an egl backend for DE-less video output via sdl?
13:02mwk: that'd be fun, but there's no sane scaling support, right?
13:02pq: mwk, that's up to the DRM driver, there could be.
13:03pq: then if you write a GL-EGL-on-KMS player, you can have it all
13:03karolherbst: hondz: I said full dmesg, your picture is quite useless
13:03mwk: hmm... there might be overlay support, it could actually be fun
13:04hondz: karolherbst: What is the command to get that ?
13:04pq: yeah, overlay support particularly
13:04pq: but why would anyone do that, when you could use Weston and run a mainstream player
13:04karolherbst: hondz: dmesg
13:04hondz: I mean, to save it to a text file
13:04karolherbst: well in theory you can run mesa on fbdev
13:05pq: karolherbst, not that it'd do any good :-)
13:05karolherbst: well :D
13:06karolherbst: I think you only need the fb device of nouveau and could get some opengl running on fbdev
13:07karolherbst: "Applications in the fbdev/DRI environment use the MiniGLX interface to choose pixel formats, create rendering contexts, etc. It's a subset of the GLX and Xlib interfaces allowing some degree of application portability between the X and X-less environments."
13:07karolherbst: pq: I am sure it isn't
13:07pq: unless, you also open the render node of nouveau, and do a glReadPixels to the fbdev mmap
13:07karolherbst: pq: http://www.mesa3d.org/fbdev-dri.html
13:07karolherbst: but I think you need a rather old mesa version for this
13:07karolherbst: could be totally broken now
13:07pq: "NOTE: this information is obsolete and will be removed at a future date"
13:08pq: karolherbst, you can't poke nouveau ioctls on a fbdev device node. It simply won't work.
13:08karolherbst: ohh I see
13:09karolherbst: yeah well, then it won't work, right
13:09karolherbst: but I think mesa uses a drm node on fbdev
13:09karolherbst: at least the site suggests so much
13:10pq: karolherbst, it's talking about Mesa 7 era.
13:10karolherbst: but yeah, this was like mesa-7 or mesa-8
13:10pq: airlied, http://www.mesa3d.org/fbdev-dri.html is here haunt you ;-)
13:10hondz: karolherbst: I am preparing it
13:11karolherbst: but in the end, nobody wants to support fbdev
13:11karolherbst: I am sure if you go to vlc with your fbdev errors, they will tell you to repair it yourself most likely
13:11karolherbst: except a vlc dev uses fbdev
13:12karolherbst: or did they remove fbdev support already?
13:13karolherbst: mhh seems like they did
13:15karolherbst: right, directfb was the name of that thing
13:15mwk: I recall directfb was yet another monster
13:16karolherbst: well directfb is a library on top of fbdev
13:16karolherbst: which application are using to do stuff
13:16mwk: apps that used "fbdev" or "fbcon" actually worked, the ones using directfb did crazy shit :p
13:17mwk: as in, fbdev crashes -> you press ^L to clean the garbage and are back to terminal; directfb crashes -> sorry, your display is stuck like that
13:17karolherbst: ... :D
13:17pq: didn't directfb have "hardware drivers" to poke the hw like the X server used to do in the 90s?
13:17karolherbst: yeah, somewhat like that
13:17mwk: and lots of apps supported both, so probably fbdev locked up shit for other people :p
13:17mwk: pq: that was svgalib iirc
13:17karolherbst: well the thing is, nothing seems to supported fbdev at least on gentoo
13:18karolherbst: don't see any fbdev USE flags
13:18pq: mwk, I recall seeing nvidia drivers in directfb...
13:18karolherbst: except in weston
13:18mwk: pq: maybe they both did it :)
13:18hondz: karolherbst: http://pastebin.com/iMuUVDXS
13:18mwk: ah, the good old times
13:18karolherbst: hondz: thanks
13:18pq: or maybe svgalib only poked the "standard" hw interfaces, while directfb went full crazy
13:18karolherbst: there ya go "nouveau 0000:0a:00.0: unknown chipset (118010a2)"
13:18mwk: when anyone and their dog poked hardware directly, because why not
13:19karolherbst: ohhh wait
13:19mwk: pq: svgalib most definitively went full crazy
13:19karolherbst: hondz: you have a laptop?
13:19mwk: doesn't mean directfb didn't :)
13:19karolherbst: hondz: or desktop with also an intel GPU?
13:19hondz: karolherbst: laptop
13:20karolherbst: then intel is your main GPU
13:20hondz: karolherbst: oh, it also has an nvidia gpu
13:20hondz: karolherbst: that's why I got confused
13:20karolherbst: but nouveau also fails to load for some bogus sutpidity
13:20karolherbst: but really, what is chipset 118010a2
13:21mwk: huh, svgalib actually still has download link for libc.so.5 :D
13:21mwk: karolherbst: GM108
13:21karolherbst: this should be supported on 4.5 right?
13:21karolherbst: yeah well
13:22karolherbst: hondz: anyway, nouveau has nothing to do in your case, cause you can only use your intel gpu
13:22hondz: karolherbst: Ok ! Thank you
13:22karolherbst: well you could use your nvidia one in X and maybe even in wayland (if support would be there for this)
13:23hondz: karolherbst: should I remove novena ?
13:24karolherbst: what is novena?
13:24hondz: kaorlherbst: I am sorry, I meant nouveau
13:24karolherbst: ahh, well it is in your kernel
13:24karolherbst: and it should also be able to turn off your nvidia gpu at least
13:25hondz: karolherbst: Any ideas about how to fix the problem ?
13:25karolherbst: but nobody really had time to look into the gm108
13:25karolherbst: somebody has to figure out the differences between gm107 and gm108
13:34pmoreau: hondz: https://bugs.freedesktop.org/show_bug.cgi?id=89558 tracks the lack of support for gm108, and has a "hacky" patch to get Nouveau to recognise the chipset.
13:34eycsigfy: Hey All, I was wondering if people had issues with Displayport and Nvidia GPU
13:34eycsigfy: 's here
13:36hondz: pmoreau: how do I apply it
13:38pmoreau: You will need to compile your own kernel, and then `git apply the_patch`. But since you most likely won’t have the same paths, probably just edit the code directly, since the patch is quite short
13:55karolherbst: huelter: but how long did it take with my branch to hang?
13:56huelter: about 5 hours
13:57karolherbst: so better than with stock nouveau?
13:57huelter: then switched to boost 0, hanged in 10 minutes
13:57karolherbst: mhh at least something
13:57huelter: now on boost 2, still no hangs
13:57huelter: but this is all anedoctal, need more testing
13:57karolherbst: I know
13:57karolherbst: but as long as it feels better and runs longer in general, it is still better than stock
13:57huelter: sure is
13:58huelter: I find odd that the voltage and power stays the same
13:58karolherbst: it is a bit, but that branch also contains more recent stuff
13:59karolherbst: sadly my GPU doesn't hang anymore at all :/
13:59karolherbst: so I can't really debug any of those
13:59karolherbst: hakzsam_: how long do you need reator today? I want to try to trigger hangs on those GPUs
13:59karolherbst: hakzsam_: and maybe need to run for several hours
14:02hakzsam_: piglit uses it actually :)
14:04karolherbst: :D I see
14:05karolherbst: now I would wish my GPU would any ports wired to it...
14:05karolherbst: then I could also test the display stuff
14:06hakzsam_: should be done in few minutes btw
14:06karolherbst: ahh good
14:06karolherbst: do you still need it after that?
14:08hakzsam_: yeah, probably later today (tonight) just to fix one thing related to images on gk208, a really minor issue
14:14pecisk: karolherbst: by Phoronix article, Fermi support incoming for re-clocking?
14:14karolherbst: hakzsam_: "Improvements"
14:14karolherbst: hakzsam_: sorry
14:15karolherbst: pecisk: "Improvements"
14:15karolherbst: well the boosting stuff is also there in fermi vbios'
14:15karolherbst: so I also took care of this at the same time
14:15karolherbst: and the big thing for fermi is memory reclocking anyway
14:16pecisk: I see
14:16pecisk: that's still not started right?
14:16karolherbst: well it is started, but not finished
14:18karolherbst: wuut, those coefficienst are soo right...
14:18karolherbst: mupuf: this just happened: https://gist.github.com/karolherbst/aa84c990c20005b89061e8dd51b58d7d
14:18karolherbst: on -3 no revolt
14:20Tom^: holy moly, just dont code yourself worn out karolherbst :P
14:20Tom^: cool changes the past few weeks
14:20karolherbst: it was fun :D
14:22karolherbst: k, it also works with the both cards in reator without any errors
14:22karolherbst: so 10 cards were fine with my old coefficients
14:22karolherbst: and 3 are perfect with the new ones :)
14:22pecisk: should GTX 850 work now?
14:22karolherbst: or do you mean gtx 850m=
14:22pecisk: yes, M
14:23karolherbst: mhh well I don't know when we want to enable memory reclocking on maxwell yet
14:23karolherbst: but it is quite easy to add
14:24pecisk: karolherbst: well, I had problem that it didn't work for me at all last time I tried with Fedora 23
14:25pecisk: now I use Bumblebee
14:25pecisk: with binary nvidia
14:25karolherbst: problem was gpu crashed?
14:25pecisk: spammed console with repeated messages and freeze
14:25pecisk: I forgot what was name
14:25pecisk: btw, one sec
14:26pecisk: I will try to flash boot on Fedora 24
14:26karolherbst: well you would need an updated branch anyway
14:26karolherbst: but maybe it will work better now
14:26karolherbst: and I've also added a nouveau parameter to increase the voltage
14:26karolherbst: there is some oddness with some GPUs
14:27pecisk: I just want to quickly check out if that message appears
14:35pecisk: karolherbst: well, it doesn't freeze, and nouveau module is loaded, so that's progress
14:35karolherbst: yeah but you used the old version still
14:36karolherbst: I think
14:36pecisk: it is possible
14:36pecisk: I get this message in dmesg
14:36pecisk: nouveau 0000:01:00.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0xffff
14:36pecisk: karolherbst: ^^
14:36pecisk: but that's all
14:36pecisk: I will run it for 20 mins, if it works, I might upgrade :)
14:44pecisk: karolherbst: what was proper way to use Optimus with nouveau without Bumblebee?
14:46pecisk: yay, it worked
14:47pecisk: seems F24 has DRI3 enabled by default on this and DRI_PRIME way works
15:15night199uk: hey, just sharing this here - maybe not useful for you guys though :-)
15:15karolherbst: mupuf: your gk208 fails to set the 0 cstate, because it maps to 796754uv, but voltage min 800000 max 1187500
15:15night199uk: this is a set of 010 editor templates for decoding nvidia vbios option roms
15:15karolherbst: night199uk: you sure you are allowed to do this?
15:15night199uk: dunno - why would i not be?
15:15karolherbst: we can*t look at it, if you aren't
15:16night199uk: i don’t work for nvidia
15:16karolherbst: because of IP
15:16karolherbst: well if you use IP of nvidia, you have a license
15:16karolherbst: and in this license stands what you are allowed to do, or not
15:16night199uk: in my country reversing is not illegal
15:16night199uk: otherwise i’m not sure
15:17karolherbst: so how did you get the data?
15:17night199uk: most of this is just what’s already reflected in open DCB 4.x specs
15:17night199uk: although the BIT tables are obviously different
15:17night199uk: i guess it’s no different to what’s in envytools
15:17karolherbst: ahh okay
15:17karolherbst: as long as it is all REed then it is fine
15:18night199uk: yeah, it’s all from REing
15:18karolherbst: but yeah, most of this is aready known I think
15:18night199uk: its basically what’s in envytools already although there are some things i have envy doesn’t, some things envy has i didn’t support yet, etc
15:18karolherbst: night199uk: did you run your vbios through nvbios?
15:18rashed: karolherbst: do you have a gm20x?
15:18karolherbst: rashed: no
15:19night199uk: and i think i already added some of the newer tables that you guys don’t parse yet
15:19night199uk: my vbios?
15:19rashed: karolherbst: hm ok, thanks though
15:19karolherbst: imirkin: there ya go :D
15:19night199uk: i’ve run a bunch through nvbios :-)
15:19night199uk: i’ve been building my own nvidia UEFI driver for 2+ years now :-)
15:20night199uk: just finishing up gtx980 support now
15:20karolherbst: ahh I see
15:20night199uk: tesla, fermi, kepler all already working
15:21night199uk: thought i’d share these as maybe they’re easier to work with for some folks than envy
15:21karolherbst: night199uk: so basically an UEFI driver to get the output working and stuff
15:21night199uk: it’s basically a replacement for the UEFI driver flashed on your card
15:22night199uk: its a replacement GOP driver
15:22karolherbst: ahh okay
15:22karolherbst: yeah, why not. Sounds a thing somebody might want to have
15:22night199uk: so just 2d basic input/output stuff
15:22night199uk: its mainly for hackintosh guys
15:22rashed: karolherbst: another quick question, do you know if DRI_PRIME is working on gm20x?
15:22karolherbst: rashed: with newest kernels and nvidia firmwares, yes
15:22night199uk: who need the custom nvidia EFI device properties injection stuff
15:23karolherbst: night199uk: ohh right
15:23night199uk: though i’m fixing quite a few bugs in the nvidia EFI drivers on the way ;-)
15:23rashed: karolherbst: ive got 4.6 rc3, mesa 11.3 and the latest firmware from linux-firmware but it doesnt seem to be working
15:23night199uk: also backporting the latest nvidia drivers to older cards
15:23night199uk: e.g. i have a fermi driver based off 84.x.x.x which is nominally the maxwell branch
15:24karolherbst: rashed: check dmesg and X log
15:25karolherbst: night199uk: well yeah, sounds great, but we are still mostly occupied with kernel and userspace stuff :) but are there any stability concerns? and how would you use the replacement in a normal desktop machine?
15:26night199uk: sorry, yeah, didn’t mean to bore with the project - was just sharing the templates back as nouveau was a really valuable reference for me
15:26rashed: karolherbst: xrandr just isnt detecting the nvidia card
15:26rashed: --listproviders only shows intel
15:26karolherbst: night199uk: well you can always improve the envytools code and add a better parsing for the unknown bits
15:26karolherbst: rashed: yeah, check the X log and dmesg then
15:27night199uk: replacement would be flashing the card right now but it will be simpler later - once i’ve finished and got this working i need to implement a couple of odds and ends to allow you to inject this driver into your bios and have it take over from the card driver
15:27karolherbst: night199uk: and I am sure projects like libreboot might find your stuff valuable
15:27night199uk: also i believe you can load the efi driver from disk
15:28night199uk: yeah, i’m on the coreboot list also :-)
15:28night199uk: ideal for hackintosh users is to reflash their cards
15:28night199uk: this way you get a GTX980 ‘Mac Edition’ since the last available ‘Mac Edition’ card was the EVGA GTX680
15:29night199uk: and yeah, i should contribute back to envytools when i’ve got some time. 010 is much easier to work with as you’re reversing as you can tweak and see results immediately rather than tweak/compile/run, etc
15:30night199uk: when i’m done with this part and tidying up i’ll try and make some time :-)
15:39mupuf: karolherbst: ah ah
15:48peteris: karolherbst: still wondering what is this - nouveau 0000:01:00.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0xffff
15:49imirkin: peteris: that's not a real problem
15:49peteris: ok good
15:49imirkin: peteris: it can be, but assuming nouveau loaded, it's fine
15:50peteris: it is loaded
15:50peteris: at least glxinfo says so :)
15:50imirkin: cool :)
16:17hakzsam_: karolherbst, done with reator?
16:17hakzsam_: oh actually I have something else to do
16:18hakzsam_: I'll ping you later ;)
16:20karolherbst: hakzsam_: yeah I am kind of done
16:25karolherbst: now the fun begins
16:35huehner: karolherbst: just noticed http://lwn.net/Articles/684415/ hwmon support patch for ina3221 apart from nouveau
16:35imirkin: yeah, i saw that one too
16:35karolherbst: ohh nice catch
16:35karolherbst: didn't see it
16:35imirkin: didn't think it was worth sharing anything though
16:36karolherbst: it is complicated
16:36karolherbst: mupuf: what do you think? Would be able to support this in any way? Or would it be just too complicated?
16:40mupuf: karolherbst: do you know anything about the history of the collaboration between hwmon and nouveau?
16:40mupuf: (or hwmon with anyone else)
16:40mupuf: lynxeye and I are done with them
16:41karolherbst: ohh what happened?
16:41mupuf: well, hwmon is a userspace API
16:41mupuf: with no support for in-kernel access
16:41mupuf: we wrote 6 versions of a patch set to allow for in-kernel users
16:41mupuf: and finally put some consistency on the code (it is the far west, every driver does what it wants)
16:41mupuf: but not mandating to rewrite all drivers
16:42mupuf: their answer first was: duplicate the code
16:42karolherbst: how that makes no sense
16:42mupuf: their second answer was to actually give some feedback
16:42mupuf: then when they could not criticize it anymore, they just said: we are too busy
16:42mupuf: and that is the situation
16:43mupuf: mjg59 also had issues getting something like that in, before we did even try
16:43mupuf: samsung also failed at reasoning with them
16:43mupuf: and when I talked to them, they did not want to recognize that a driver not maintained by them could expose the hwmon interface
16:44mupuf: although they did check at some point and agreed with it
16:44mupuf:is not an ass, when an interface is good-enough, one uses it
16:44mupuf: but this "only userspace API" bullshit really killed me!
16:45mupuf: putting the driver in FUC was a way to avoid this
16:45mupuf: but in the end, no-one fucking cares, so, you just added it to the kernel :p
16:45karolherbst: well we expose some hwmon stuff :D
16:46mupuf: all who knew the situation shut up and let it happen
16:46mupuf: and all who did not know where completely happy with this
16:46mupuf: we expose more things than any other driver :p
16:46mupuf: and I like it this way
16:47mupuf: and we can definitely expose more
16:47mupuf: like the warnings about overtemperature and over current
16:47mupuf: but ... one thing at a time
16:47mjg59: Yeah that was somewhat infuriating
16:47karolherbst: mupuf: saw my patches about the min/max volt thing?
16:48karolherbst: ohh did I send those out?
16:48karolherbst: ohh seems like I didn't
16:49Tom^: karolherbst: https://lists.freedesktop.org/archives/nouveau/2016-April/024698.html
16:49karolherbst: not this
16:49karolherbst: this: https://github.com/karolherbst/nouveau/commit/eab96c9d86c4d051ede66087f70704d3ba826925
16:54mupuf: looks good
18:46karolherbst: that SR2 port is friggin bad
18:46karolherbst: feels like playing on a fx 5700 or something
18:47towo`: that's known, and the win version is not many better
18:47towo`: so taht version can't even be better, since it is a EON "port
18:55karolherbst: yeah but even with nvidia the performance is shit
18:55karolherbst: lowest settings, 40 fps on full hd
18:55karolherbst: on a 770m...
18:56karolherbst: ohh "Due to issues surrounding the original engine, some stuttering may be observed during gameplay. These types of streaming issues may be improved with faster HDD units"
18:56karolherbst: yeah I noticed that
19:08Celti: SR2 played smooth for me (on blob, GT745M) until my laptop started overheating.
19:08Celti: Then again, I've got everything on an SSD.
19:18karolherbst: yeah, maybe it does a crapload of io
19:19hakzsam_: imirkin, fyi, Passed: 2061/2061 (100.0%) (ssbo on gk208)
19:19hakzsam_: I would say, it's not that bad :p
19:20Yoshimo: understatement of the day
19:42mupuf: hakzsam_: yeepee
19:43imirkin_: hakzsam_: thanks for double-checking
19:49Dezponia: karolherbst: SR2 is considered one of the worst consol ports on Windows as well so not surprising that the GNU/Linux port is lacking :) Still was my favourite SR game though :)
19:49karolherbst: I liked 4 too, but it was the only one I played fully
19:49karolherbst: started to play 2 but well the performance..
19:49karolherbst: even no fun on nvidia
19:50Dezponia: karolherbst: In other news you've been busy poking at kepler I've heard. I should throw those patches on the TITAN Black and see if anythings different
19:51karolherbst: it will be different
19:52Dezponia: Guess I've lost the Heaven score record then until I get around to running it again :P
19:53mupuf: Dezponia: with a titan black... no wonder!
19:53Dezponia: mupuf: ?
19:54karolherbst: the black isn't so fast
19:55Dezponia: Clearly I need 3 more of them for SLI
19:55karolherbst: or 2 Titan Z :)
19:56karolherbst: but the Titan Black is just like a 780 ti only with some double precision magic
19:56Dezponia: Nah the TITAN Z just has 2 "regular" TITAN chips, the Black has the full blown core. Basically the difference between the 780 and 780Ti
19:56Dezponia: So clearly dual TITAN Z's are for casuals and we cant have that
19:56karolherbst: a titan z still has the same anount of cores
19:57karolherbst: but just doubled
19:57Dezponia: Once you go Black you never go back et cetc
19:57mupuf: Dezponia: AHAH
19:57Dezponia: karolherbst: The Z is also lower clocked by default by quite a bit
19:57karolherbst: that's the smallest problem
19:57karolherbst: just get better cooling
19:57karolherbst: amateur :p
19:58Dezponia: Well regardless I assume someone has broken my record with these new patches floating around since I haven't tested them yet :P
19:59karolherbst: not really
19:59karolherbst: a R9 fury maybe
20:00Dezponia: I'm likely to go full AMD with my next build provided ZEN works out nicely
20:01karolherbst: yeah well
20:01karolherbst: if you say so :p
20:01karolherbst: I just tell you, that in some cases nouveau runs more efficiently than radeon do
20:02Dezponia: karolherbst: Well its not like the titan card is going anywhere :P
20:02Dezponia: I still have my GTX680 4GB edition laying around until i figure out something to use it for
20:03karolherbst: well a 780 ti running nouveua can beat a r9 fury in some benchmarks running radeon, which is pretty intense
20:03karolherbst: I know one thing
20:03Dezponia: Thats promising though :)
20:03karolherbst: stability testing
20:03imirkin_: amd stack is supported by full-time developers with access to docs
20:04karolherbst: yeah, exactly
20:04karolherbst: that's what bothers me most
20:04imirkin_: nouveau's 3d component is maintained by volunteers without docs
20:04imirkin_: so... go with amd.
20:04karolherbst: yeah the support will be better
20:04karolherbst: and general experience
20:04karolherbst: and stability
20:04imirkin_: a lot more stable, a lot more responsive, etc
20:05karolherbst: right :/
20:05karolherbst: but I think we will figure out most of the issues at some point
20:07imirkin_: maybe, maybe not
20:08imirkin_: a bunch of tesla issues remain
20:08imirkin_: look for 406040
20:08imirkin_: nfc what causes it
20:08imirkin_: happens a lot to some people, rarely to others
20:10karolherbst: what is 406040?
20:11imirkin_: an error :)
20:19Dezponia: Clearly we need to buy a ton of TESLA cards and send them your way
20:19Dezponia: For science
20:20imirkin_: to be clear, i'm talking about the tesla gpu family
20:20imirkin_: not the tesla marketing term
20:20Dezponia: Either or
20:20Dezponia: Same difference
20:20Dezponia: Lets just buy everything with the name TESLA just to be sure. How many of the cars do you need?
20:21mupuf: Dezponia: that's the thing, we do not have an infinite bandwidth
20:21mupuf: and GPUs are rarely the issue
20:21mupuf: although, for the GK110, it was :D
20:21mupuf: but hey, it is apparently stuck in the Finnish customs now
20:21imirkin_: well, GK208 has the same isa
20:22mupuf: imirkin_: yep, but what about performance?
20:22mupuf: GK208 is a joke
20:22imirkin_: about the same? :p
20:22imirkin_: perf-per-watt... dunno
20:22imirkin_: it's like 25W
20:22imirkin_: vs the GK110 which is 25MW
20:23Dezponia: I think its a great heater in the winter
20:23mupuf: hopefully, I will have it by the week end!
20:23mupuf: that leaves one day for customs to clear the package, we'll see!
20:25Dezponia: I just tried out Amazons "Expedited Shipping" option yesterday
20:25Dezponia: Ordered after lunch yesterday, got it before lunch today. From england to sweden
20:26Dezponia: Truth in advertising I guess
20:26hakzsam_: imirkin_, have fun! images for gk208 is on mesa-dev :)
20:27hakzsam_: but it's the same, wtvr
20:27Dezponia: I should know
20:31imirkin_: hakzsam_: i think emitSUSTGx is a bit broken for SUSTGB... but that's never used right?
20:32hakzsam_: imirkin_, it's not used yeah
20:32hakzsam_: to be honest I think it's also broken on nvc0
20:32hakzsam_: but I double-checked on gk110, should be fine
20:34hakzsam_: imirkin_, okay for the assert :)
20:44hakzsam_: imirkin_, oh yeah, 100% of Rb :)
20:45imirkin_: i tend not to be _that_ picky
20:46imirkin_: and you've gotten a lot better at your submissions :)
20:46hakzsam_: yeah, thanks
20:55huelter: karolherbst: managed to lockup again with pstate 07 and boost 2, attached a different kernel log with new nouveau info https://bugs.freedesktop.org/show_bug.cgi?id=95031
20:56karolherbst: huelter: with the raised voltage?
20:56huelter: not yet, but found the new info interesting
21:00huelter: gonna up the voltage now, reset
21:04huelter: so far so good, I hope I don't get housefires :P
21:04huelter: 4w increase with 100mV
21:05karolherbst: same clock?
21:05karolherbst: or a bit lower
21:06huelter: boost 0, AC: core 324 MHz memory 648 MHz
21:06karolherbst: ohh 07 pstate
21:06huelter: yep, same clock
21:06karolherbst: the boost has no effect as long the pstate clocks are below the boost one
21:06huelter: I see, no need to change it on 07 then
21:07karolherbst: and usualy also no need on 0a
21:08huelter: what would happen if I increased 1V? Does it respect vbios maximum?
21:09karolherbst: yeah, it does
21:09karolherbst: if you set 1V, you will still be able to reclock
21:09karolherbst: because every cstate defaults to info.max
21:09karolherbst: in that case
21:10huelter: can I read the maximum somewhere?
21:10huelter: without trying to set it past fisrt
21:11karolherbst: well that's the patch, maybe you understand how it works: https://github.com/karolherbst/nouveau/commit/3c08aa9a70ceb0cbe3f3d0a193a2b48d2fd89d57 :D
21:11airlied: pq: miniglx ftw!
21:11karolherbst: ohh wait
21:12karolherbst: I don't respect info.min info.max in the entries itself...
21:12karolherbst: yeah, maybe it is better this way, have to think about it
21:12karolherbst: well the max limits are still not touched by the offset
21:12karolherbst: for obvious reasons
21:16imirkin_: hakzsam_: probably do the same thing for IMUL for the GM107 emitter?
21:19imirkin_: hakzsam_: perhaps also check out whether this is an issue for other integer ops?
21:19imirkin_: hakzsam_: perhaps we really do need to adjust what we allow for integer short immediates... hopefully not.
21:43imirkin_: RSpliet: looks like someone with a G200 having trouble calculating MR's
21:44imirkin_: RSpliet: https://bugs.freedesktop.org/show_bug.cgi?id=95044
21:45hakzsam_: imirkin_, thanks for the hints, I'll have a look later, I'm dealing with gk104
21:46imirkin_: hakzsam_: kk
22:03RSpliet: imirkin_: thanks for the notification, I think that's the same person that karolherbst kicked to the mailing list on the phoronix forums
22:04karolherbst: ahh right
22:04RSpliet: okay, battered
22:04karolherbst: yeah I have a slight noticable headache, and not really that much able to think straight :)
23:37biker_rat: Much better than March 29 edition.
23:38karolherbst: how so? :)
23:39biker_rat: Bioshock & Saints Row 4 now playable with Karol's v4 boost on GTX 660ti
23:39biker_rat: Not doable 3 weeks ago.
23:40biker_rat: I am going to try metro last light in 5 minutes.
23:40karolherbst: you shouldn't do that
23:40karolherbst: metro triggers some weird out of memory issue
23:41karolherbst: you system will hang
23:41karolherbst: and die
23:41karolherbst: or start killing random processes
23:43biker_rat: Will it hurt anything or can I just cold restart? This is a testing machine. I installed it 3 months ago just to fool with games.
23:44karolherbst: you can wait a few minutes
23:44karolherbst: then the kernel should start killing everything
23:44karolherbst: disable swap
23:44karolherbst: this helps
23:44karolherbst: because the kernel doesn't even try to swap stuff out
23:44karolherbst: I once run metro with zram
23:44karolherbst: and had a memory usage of above 40GB with that game...
23:44karolherbst: a lot of fun
23:45biker_rat: I don't believe in swap.
23:47biker_rat: okay I am pressing play
23:47biker_rat: it quits instantly
23:47karolherbst: I think it only happens when you start
23:47karolherbst: it shouldn't though
23:51biker_rat: I'll try metro 2033 now