00:04 fdobridge: <m​arysaka> very unimportant question but I have a fault with my GTX 960 and 970 respectively when nouveau kernel driver init them
00:04 fdobridge: <m​arysaka> ``bus: MMIO write of 8000010e FAULT at 10eb14 [ PRIVRING ]``
00:04 fdobridge: <m​arysaka> I have been having that for quite some years but it seems to not affect anything anyway, any ideas of what this might be?
12:36 fodasso: Hello. Sorry if this is the wrong place to ask, but the other day you told me it would be better to use the drm-next kernel branch to get the best results with nvk
12:37 fodasso: However, I can't find that branch on git.kernel.org
12:37 fodasso: Besides, I'm segfaulting on mesa-22.3.0 (even with standard nouveau driver, without nvk) and unable to start sway. Is it a known issue? I can't find anything related to this on my distro's bug tracker.
12:37 hell__: drm-next would probably be in the freedesktop git
12:42 fodasso: https://cgit.freedesktop.org/?q=drm-next "No results found"
12:45 hell__: https://cgit.freedesktop.org/drm/drm
12:45 hell__: second line, in the "Branch" section
12:45 hell__: or Ctrl-F (web browser's in-page search) for "drm-next"
13:51 fodasso: Great! tyvm
18:37 fdobridge: <m​henning> @jekstrand the instruction latencies you're using in nak look a little funky to me. It's possible the ones here are more accurate: https://arxiv.org/pdf/1903.07486.pdf#section.4.1
18:53 RSpliet: mhenning: I remember looking at instruction scheduling as a little experiment many moons ago. My conclusion back then was that nouveau was still in such a state that the only thing that mattered was "memory accesses are expensive, the rest is cheap. And aim for dual issue on those weird Kepler/Maxwell cards"
19:00 mhenning: sure. just pointing out a resource, in case it's helpful
19:01 RSpliet: Thanks!
19:10 fdobridge: <j​ekstrand> Thanks! That looks like a gold mine!
19:15 fdobridge: <m​henning> sure! the same group also published similar documents for some of the older gpus (going back to kepler I think), but the one I linked is probably the most useful right now
19:33 fdobridge: <j​ekstrand> Yeah. I've got the contacts at NVIDIA that I can probably get real numbers.
19:34 fdobridge: <j​ekstrand> I just need to ask
19:38 airlied: pretty sure they consider them super secret
19:43 fdobridge: <🌺​ ¿butterflies? 🌸> or on Ampere (sm_86) onwards