03:44 mangodev[d]: mhenning[d]: didn't even see the ping, but i already have mine patched :P
03:44 mangodev[d]: i keep up-to-date on all MRs relevant to my system
03:44 mangodev[d]: idk which patch did it, but holy hell is discord laggy
03:45 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1413369354556538980/image.png?ex=68bbae49&is=68ba5cc9&hm=3d6dadab00bbc9364246da71d81e9d960bc3c08c9015ee5737f9e0d6d854ceaf&
03:45 mangodev[d]: and the kitty bug is still there
03:48 mangodev[d]: i think the discord lag may just be from the new update
03:48 mangodev[d]: they may've just screwed around with the stylesheets a bunch, and made it laggier in the process
03:48 mangodev[d]: my pc is pretty sensitive to css lag
03:50 mhenning[d]: okay, well if it is an nvk regression let us know
04:04 mangodev[d]: mhenning[d]: update: significantly better, but still not fixed completely
04:06 mangodev[d]: easy test site: the youtube homepage
04:06 mangodev[d]: especially whatever on earth shorts do
04:06 mangodev[d]: NVK does *not* like anything to do with shorts
04:06 mangodev[d]: i think it's a mix of `border-radius` and `backdrop-filter` messing things up, since the page (especially the shorts section that's very prone to flicker) gets affected the most
04:09 mangodev[d]: i'll test trello too, as it was also pretty bad last time i checked
04:11 mangodev[d]: still flickers every now and then
04:11 mangodev[d]: hard to tell because i can't find that one issue that had a good amount of scroll space
04:12 mangodev[d]: wait no that's why
04:12 mangodev[d]: filters :|
04:13 mangodev[d]: interesting
04:13 mangodev[d]: it passively flickers
04:14 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1413376652146245632/Screencast_20250904_231343.webp?ex=68bbb515&is=68ba6395&hm=32c110fd8c03a2d96569af7158a6b9b658488020d875f4b72d4381b5a659c31b&
04:14 mangodev[d]: don't even have to scroll
04:14 mangodev[d]: strange indeed
04:20 mangodev[d]: something also interesting is that chromium doesn't have this issue
04:21 mangodev[d]: rare that it is the one *not* bugging out on nvidia, given proprietary has always been the *other* way around
04:42 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1413383641202954250/Screencast_20250904_233919.webp?ex=68bbbb97&is=68ba6a17&hm=2d3e2928cdcca1251f763aefbf8c8006d5b27f1d9382ddcb910f1fd8785d64b8&
04:42 mangodev[d]: user pages have really bad flickering too
04:42 mangodev[d]: couldn't capture it super well on video, but some parts would be "stuck" on screen and continue to flicker past when they should be visible
04:42 mangodev[d]: maybe repaints are failing to apply to all buffers correctly?
10:14 snowycoder[d]: If anyone wants to take a look at MR!37108 (cross-block scheduling) it's ready, I'm running dEQP
11:19 phomes_[d]: snowycoder[d]: I reran my tests. All looks good. There was a surprising regression in AoE4 but I can see that it comes from a game update that happened between I recorded the baseline and tested your MR
11:22 snowycoder[d]: phomes_[d]: Thanks, It doesn't gain too much performance, but it should never regress, it computes strictly lower delays
13:52 gfxstrand[d]: mangodev[d]: Is that with Firefox or Chrome?
15:13 gfxstrand[d]: And do we have an issue filed for it?
20:02 karolherbst[d]: fyi, nak fails to build on 25.2 with rust 1.78: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37198
20:20 phomes_[d]: 🚀 shared memory QMD is getting merged. And we got the conditional render performance improvement earlier today too
20:21 phomes_[d]: I was looking for games using cond render to test with but I don't think I have any installed
20:22 mhenning[d]: Yeah, I suspect the cond render one is pretty niche
20:27 snowycoder[d]: What happened with the occupancy talk of some day ago? That seemed like a very nice performance win.
20:31 karolherbst[d]: that's the shared memory QMD
20:31 karolherbst[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37135#note_3082221
20:32 karolherbst[d]: it's not often that a 2x perf bump in a micro benchmark also translates to significant perf gains in real world usage
20:33 karolherbst[d]: but we can improve things even further, we just need to model occupancy even better 😄
23:41 airlied[d]: karolherbst[d]: might want to bookmark https://www.modular.com/blog/matrix-multiplication-on-nvidias-blackwell-part-2-using-hardware-features-to-optimize-matmul for future reading 🙂
23:42 karolherbst[d]: we actually got docs on it all
23:43 airlied[d]: It's a good read on how to use the hardware effectiv
23:44 mohamexiety[d]: oh wow this is comprehensive.. but this looks to be talking about DC blackwell so I wonder how much of it can be applicable here
23:44 mohamexiety[d]: really looks like an interesting read though, thanks!
23:48 karolherbst[d]: right... though we do have other docs on that as well
23:48 karolherbst[d]: though not specific to coop matrices
23:49 karolherbst[d]: maybe I should make use of the information there because it does explain shared memory bank conflicts 😄
23:50 karolherbst[d]: anyway, the blog post isn't correct in regards to shared memory bank conflicts
23:52 mohamexiety[d]: ooh? how so
23:53 karolherbst[d]: no bank conflicts within a warp
23:56 karolherbst[d]: well it's a bit more detailed then that
23:56 karolherbst[d]: *than
23:59 karolherbst[d]: or maybe "isn't correct" is a bit too harsh, but it's not as detailed and precise as it should be