10:44thican: Hello guys, I have a question about Vulkan support: does it work (or will) with Nvidia GPU?
10:45mupuf: thican: yes, but not with nouveau for quite some time
10:45thican: ok, thanks for the answer :-)
10:45thican: what about Nvidia's propri drivers?
10:48thican: Another question, please, will it work with any Nvidia architecture, like old GPU?
10:48RSpliet: thican: no, iirc Kepler and newer only
10:58thican: ok, thanks. :-)
20:03mwk: wheee, got a green test run with NV10
20:04mwk: I wonder how much more hammering NV15 would need...
20:07pmoreau: a green test? output a uniform green color on a screen?
20:10mwk: pmoreau: green as in correct
20:10mwk: ie. if I can simulate a method on NV4, I can simulate it on NV10 as well
20:10pmoreau: Ah! Nice :-)
20:11mwk: and I've got most of state setting methods already
20:12karolherbst: mwk: I guess doing the same you do for those old cards, would take 10 years for modern kepler cards?
20:13karolherbst: or isn't it _that_ complex actually
20:13mwk: yeah, it'd be ridiculously complex
20:14mwk: I mean, state setting is easy
20:14mwk: but the more interesting tests would be quite hellish
20:15mwk: Kepler has a ridiculous degree of parallelism, and the current hwtest model relies on determinism
20:16mwk: that's... not a good combination
20:17mwk: though a lot of parts can be tested in isolation
20:18mwk: I've covered a good chunk of Tesla ISA, for instance
20:20karolherbst: ohh I see
20:20karolherbst: so tesla would be actually manageable
20:20karolherbst: and with fermi it goes all south
20:21Yoshimo: what benefits would that bring if you cover a card series completely compared to lets say the state of maxwell?
20:58mykon: any goodies in kernel 4.8?
20:58mykon: I am about to do the jump
20:59imirkin_: mykon: which gpu? and which jump are you referring to?
21:00mupuf: karolherbst: ok, I have a lead for the frequency thing
21:00mupuf: Will test it later
21:08karolherbst: :) very good
21:09karolherbst: mupuf: mind going through my patches again? even if I addressed some of your comments in a half-assed way
21:09karolherbst: mykon: depending on your gpu, but the best goodies are in 4.10 :p
21:14mykon: karolherbst: really? such as?
21:15mykon: 4.10 possibly after the summer
21:16karolherbst: mykon: LED support of course
21:17mykon: wut? the leds in card?
21:17karolherbst: no idea when it started, but high end cards have LEDs
21:20mykon: and how's that useful for the 3D game performance??
21:21mykon: or the fan running at 100%
21:21karolherbst: everybody knows that LEDs make your rig 30% faster :(
21:21karolherbst: mykon: ohh, you have fan issues?+
21:21karolherbst: mupuf: is working on that
21:21mykon: well, they're running at full throttle
21:21mykon: ALL the time
21:21karolherbst: mykon: fermi card?
21:21karolherbst: or kepler?
21:21mykon: no power management whatsoever
21:22karolherbst: well, there are also tons of reclocking improvements for fermi and newer in 4.10
21:22karolherbst: but the LED stuff is more important :p
21:23mykon: the blinking leds that you can't see because of the PC case??
21:23karolherbst: so if you are not unlucky
21:23karolherbst: you get fully working reclocking with 4.10
21:23karolherbst: yeah, I think so
21:24karolherbst: mykon: we have no 780 ti cards running at max clocks pretty stable now with the newest stuff
21:24mykon: I think I've seen mine blinking at a times
21:24karolherbst: what gpu?
21:24karolherbst: sounds like 680
21:25mykon: did you mean 'we have now ...' or 'we have NO ...' ?
21:26karolherbst: either my spelling is bad or my keyboard messes sometimes up... happens quite often to me
21:28karolherbst: mykon: well, even with the 4.10 stuff you would have to manually set the clock states, but I also have a protoype for dynamic reclocking, just have to clean it up and fix all remaining issues
21:35mwk: imirkin_: the endianness switch seems to have a lot of funny effects
21:36mwk: the first test that blew up in my face was setting pattern colors
21:36mwk: eg. PATTERN_Y8 method has pixel (0,0) in bits 0-7 for LE, 24-31 for BE
21:37mwk: likewise, PATTERN_MONO_BITMAP is affected
21:37imirkin_: mwk: well remember that it byteswaps all pushbuf input
21:37imirkin_: mwk: and, i think, register accesses
21:37imirkin_: which makes it a lot more fun to attempt to analyze
21:38imirkin_: mwk: oh, and you ahve to set a bit in the context object as well
21:38imirkin_: word 0 of the object needs to be set to 0x00080000
21:39mwk: yeah, I'm talking about the context object bit
21:39imirkin_: oh. but there's another bit.
21:39mwk: I haven't touched the PMC switch
21:39imirkin_: yeah ok
21:39imirkin_: flipping both of them together is the usual thing, i suspect :)
21:39mwk: I think the PMC switch is simple, it swaps register accesses and FIFO pushbuf words
21:40mwk: the PGRAPH bit is much more interesting
21:40imirkin_: i'd also be curious what effect it has on reads from GART, if any
21:40imirkin_: [i assume none]
21:40mwk: I guess it depends
21:41mwk: if you're reading byte data, probably none
21:41mwk: but if you're reading R5G6B5... yeah, fun
21:42imirkin_: right - and i think that the various read services auto-byte-swap ... sometimes? but you can also tell them whether they're reading bytes or shorts or words
21:42imirkin_: but i'm not sure how
21:42imirkin_: i think it's actually documented, either for g80+ or for pre-g80
21:43imirkin_: either way, all that stuff gave me a headache =/
21:43mwk: yeah, it's going to be a mess
21:43mwk: we'll see once I get to the rop tests on NV11
21:43mwk: might take a while
21:45imirkin_: i think the idea is that when all the bits are flipped, the code which uses this stuff doesn't change
21:45imirkin_: so the pattern is flipped because the value will be in a pushbuf and that will get byteswapped
21:45mwk: yeah, that seems to be the general idea
21:46mwk: I wonder how that works on G80+, where shaders can do arbitrary-sized loads/stores
21:47mwk: ugh, NV15_SWZSURF
21:47mwk: wtf happened to this poor object
21:48mwk: it seems to be completely broken, whatever format you stuff into it, it comes out as invalid
21:48imirkin_: sounds like you're missing a bit of setup
21:48imirkin_: check what nouveau does?
21:48imirkin_: although i don't know that it was ever tested on a NV15
21:48mwk: nouveau skips this object on NV1x because it doesn't work
21:49imirkin_: oh :)
21:49imirkin_: see - problem solved! :)
21:49mwk: and there's no way I could be missing a bif of setup...
21:50mwk: ... I hope
21:54mwk: ugh, of all the methods on this thing, *nop* is giving me trouble
21:55mwk: "PGRAPH locked up !"
21:55mwk: but whyyyyy
22:08The_New_Guy: hey guys!
22:08The_New_Guy: I am in desperate need for help, I am trying to install Linux Kali with having a geforce gtx 1070
22:09The_New_Guy: but the nouveau drive has a really hard time booting up
22:09The_New_Guy: the deal is, my machine cannot connect to the internet
22:09imirkin_: try booting with nouveau.modeset=0
22:10The_New_Guy: let me google how to implement it
22:34The_New_Guy: GRUB_CMDLINE_LINUX_DEFAULT="nouveau.modeset=0" did not do the trick
22:35imirkin_: then nouveau has nothing to do with your issues.
22:35The_New_Guy: i have no idea what to do, i have a gtx1070 installed and i get no linux to display anything, except for the latest kali weekly
22:36The_New_Guy: i get a slowly establishing login window with the terminal popping in all the time
22:36The_New_Guy: it has something to do with the graphics
22:36The_New_Guy: its not necessarily nouveau, but I have no idea what to do
22:42imirkin_: sounds like you need help from your distro
22:43mwk: wtf, I actually managed to find some new unknown methods