00:01 TimurTabi: Can you run the same test with OpenRM?
00:34 fdobridge: <E​sdras Tarsis> TimurTabi: ~4300 FPS
00:34 TimurTabi: Seriously? Nouveau is faster?
00:36 fdobridge: <E​sdras Tarsis> Apparently yes, I'm testing it on sway
00:36 fdobridge: <E​sdras Tarsis> glxgears is a black window on xorg with nouveau gsp
00:47 fdobridge: <🌺​ ¿butterflies? 🌸> glxgears is not a benchmark
00:49 fdobridge: <E​sdras Tarsis> I know, but it's the only thing that works on nouveau gsp without gpu hang
00:50 airlied: does openarena not work?
00:52 DodoGTA: TimurTabi: Is your PR in Ben's nouveau repository still relevant?
00:58 TimurTabi: Yes
00:58 fdobridge: <E​sdras Tarsis> airlied: I'll test, but even super tux kart didn't work
01:01 TimurTabi: DodoGTA: I've made a few comments over the past several months. My only real concern is this one: https://gitlab.freedesktop.org/skeggsb/nouveau/-/commit/2541626cfb794e57ba0575a6920826f591f7ced0#note_1700578
01:01 TimurTabi: I don't know how to fix that one.
01:01 TimurTabi: It prevents me from enabling IOMMO on my test systems.
01:03 DodoGTA: TimurTabi: I always have IOMMU enabled on my system so that's an issue for me too
01:03 TimurTabi: I think the problem shows up now because the GSP-RM binaries are so large. This is especially try of the debug versions, which I use. They can be 90MB in size.
01:04 TimurTabi: s/try/true
01:57 fdobridge: <E​sdras Tarsis> airlied: Open Arena works, I think the game is capped at 91 FPS
01:57 fdobridge: <E​sdras Tarsis> both nvidia-open and nouveau gsp get 91 FPS in game
02:00 fdobridge: <k​arolherbst🐧🦀> :3
02:00 fdobridge: <k​arolherbst🐧🦀> no, we are just that good 😛
04:07 fdobridge: <E​sdras Tarsis> nouveau-gsp glmark2 Score: 12400
04:07 fdobridge: <E​sdras Tarsis> OpenRM (nvidia-open-gpu-kernel-modules) glmark2 Score: 6962
04:07 fdobridge: <E​sdras Tarsis> https://cdn.discordapp.com/attachments/1034184951790305330/1077803810736574474/glmark2-nvidia.txt
04:07 fdobridge: <E​sdras Tarsis> TimurTabi: amazing results
11:30 fdobridge: <J​., Echo (she) 🇱🇹> Now should I try NVIDIA closed-source kernel driver? 🐸
14:18 TimurTabi: Sure, but make sure you run it in monolithic mode (i.e. without GSP).
14:37 fdobridge: <J​., Echo (she) 🇱🇹> I weirdly only get a score of 2539 (maybe XWayland overhead?)
14:38 fdobridge: <E​sdras Tarsis> Test on xorg
17:42 Rayyan: what is fdobridge bridged to?
17:42 fdobridge: <J​., Echo (she) 🇱🇹> Discord
17:44 Rayyan: nouveau has a discord? since when :D
17:47 karolherbst: wanna join the fun?
18:04 TimurTabi: I'd like to join
18:05 karolherbst: https://discord.gg/PtkezKSAXd
19:20 fdobridge: <E​sdras Tarsis> @ttabi What do these switch case values mean? Is there similar code in openrm? (welcome to discord, btw)
20:44 TimurTabi: Esdras: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/main/src/common/sdk/nvidia/inc/ctrl/ctrl0073/ctrl0073specific.h#L1078
20:45 TimurTabi: After Ben's code is upstream, depending on which OpenRM header files he includes with it, I will be posting patches that replace magic numbers with macros from those header files.
20:48 fdobridge: <E​sdras Tarsis> ty
21:38 fdobridge: <E​sdras Tarsis> Is this the code equivalent to the switch case nouveau disp in openrm: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/main/src/nvidia-modeset/src/nvkms-rm.c#L486 ?
22:08 TimurTabi: Looks like it, but I can't say for sure. I'm not familiar at all with the SOR stuff, whatever that is. I don't even remember how I figured out the fix.
22:09 TimurTabi: I'm sure Ben will incorporate the fix at some point.