21:44fdobridge: <gfxstrand> Just run the Vulkan CTS with NVK on Maxwell
21:44fdobridge: <gfxstrand> It'll freeze up about 2 times out of 3
21:45fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Is this because of the kernel driver? 🍩
21:46fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I mean the freeze when running CTS
21:47fdobridge: <gfxstrand> Combination
21:47fdobridge: <Mohamexiety> yeh pascal too, sometimes a few weird things happen
21:47fdobridge: <gfxstrand> NVK has bugs which cause hangs and the kernel's hang recovery isn't great.
21:47fdobridge: <Mohamexiety> most stable NVK experience I saw as Turing. I think Ampere should be good too
21:48fdobridge: <Mohamexiety> most stable NVK experience I saw was Turing. I think Ampere should be good too (edited)
21:49fdobridge: <karolherbst🐧🦀> are you on 6.3 by any chance?
21:49fdobridge: <karolherbst🐧🦀> ahh yeah.. if you like your fs, don't run the CTS on 6.3 🙃
21:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Is the recovery better than AMDGPU driver's though?
21:50fdobridge: <gfxstrand> I'm on something ancient
21:51fdobridge: <gfxstrand> Some branch from skeggsb based on 6.1
21:51fdobridge: <gfxstrand> I should probably update
21:51fdobridge: <gfxstrand> hehe
21:52fdobridge: <karolherbst🐧🦀> but yeah.. pre turing is kinda a mess, no idea why
21:55fdobridge: <karolherbst🐧🦀> I think we problaby just don't reap the channel often enough...
21:56fdobridge: <karolherbst🐧🦀> like this `RT_WIDTH_OVERRUN` error looks super fatal
21:58fdobridge: <karolherbst🐧🦀> but I suspect we also send a bunch of trash up to the GPU 🙃
21:59fdobridge: <karolherbst🐧🦀> or have some annoying race on the kernel side
22:00fdobridge: <karolherbst🐧🦀> but I don't ask for any issue, and nvk won't trigger the one I want to see
22:01fdobridge: <karolherbst🐧🦀> maybe I shouldn't bother with verifying that https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19543 fixes some kind of freezes
22:01fdobridge: <karolherbst🐧🦀> but I've seen the kinds where a game/app keeps spinning and the desktop freezes until the app gives up
22:03fdobridge: <karolherbst🐧🦀> anyway.. running with less threads makes it more reliable afaik
22:04fdobridge: <karolherbst🐧🦀> but fixing races on the kernel side is annoying 😢