2026-04-13T09:55:26 <.gonzo.> @Capo is there an issue about normal map performance on windows versus linux? sophie and i testing it and her performance with normal maps on is like my performance with normal maps completely off. with normal maps on the game is nearly unplayable flying around. constant hitching and lag spikes. with it off you still get occasional hitches, but they only last a few ms instead of full seconds. seems to be a major 2026-04-13T09:55:26 difference between OS builds, because our hardware isn't that much different. the hardware difference is 4090/7950x3d (hers) vs. 9070xt/7800x3d (mine) 2026-04-13T09:56:03 <.gonzo.> exact same settings.cfg configuration also 2026-04-13T09:56:08 <.gonzo.> and mod configuration 2026-04-13T09:56:21 <.gonzo.> except she's at 4k resolution i guess and i'm at 1440p 2026-04-13T09:56:58 <.gonzo.> maybe ati vs nvidia difference? 2026-04-13T09:57:01 You'll probably have to check up with AON3 on that because I'm not aware of anything that could lead to differences 2026-04-13T09:57:19 Could be driver-related, I guess 2026-04-13T10:00:05 <.gonzo.> i'm tempted to install linux on this thing just to rule out hardware differences lol 2026-04-13T10:35:41 yeah not gonna lie I had no idea that cell preload stutter could be that bad. :sadcat: 2026-04-13T10:47:56 Good to check if the normal maps have mipmaps too, that was a major source of stutter for me until I realized several normal map mods were made without them 2026-04-13T11:16:18 I know some other guy as well mentioned that amd cards specifically have extra stutter with normal maps so he avoids using them. For me it's pretty smooth flying at 4k speed on nvidia gpu/linux. I guess if it's driver related not really sure what can be done there. I do think it's worth exploring probably. 🤷‍♀️ 2026-04-13T13:06:45 The contribution guidelines mention "Start a discussion" for planning more sweeping changes/features - where is the best place to "start a discussion?" 2026-04-13T13:10:00 Usually Discord/Matrix/IRC (i.e., somewhere around here), rarely forums but they're less used 2026-04-13T13:10:44 Appreciated, would this be the right channel for that then? Or am I missing one (as in, maybe I don't have it selected in the channel list)? 2026-04-13T13:37:50 We have #development-lua , probably best used for that 2026-04-13T13:38:07 If you want a separated discussion, you can just make a thread (+ button the left of the message box) 2026-04-13T13:44:26 Thanks, Urm - never noticed the option to create a thread off a standard text-channel in that + options 2026-04-13T14:31:18 Assuming it's not just the cache hit rate being terrible due to a lack of mipmaps, normal maps should be nearly free on any hardware capable of running OpenMW at a reasonable framerate in the first place. The driver might affect the size of the nearly, but I wouldn't expect there to be a measurable performance difference if you looked at the framerate or anything that actually affects the user's experience. 2026-04-13T14:32:13 If the tests are comparing different hardware on different operating systems, then there are far too many variables to start blaming anything in particular. 2026-04-13T14:33:55 <.gonzo.> yeah i agree 2026-04-13T14:34:18 <.gonzo.> please let me know how i can help if at all, because it's pretty uhh... well i mean we can capture video of our experiences if you'd like 2026-04-13T14:34:32 <.gonzo.> but it's not good 2026-04-13T14:43:08 So on the computer where it's slow, disabling auto use object normal maps makes it not slow? 2026-04-13T14:50:18 <.gonzo.> yeah it's a wild difference 2026-04-13T14:50:55 <.gonzo.> i also tried auto use terrain normal maps but that didn't have really any impact 2026-04-13T14:59:38 <.gonzo.> you can see the difference here 2026-04-13T14:59:38 <.gonzo.> https://cdn.discordapp.com/attachments/260443767482482688/1493264394707013794/2026-04-13_08-52-40.mp4?ex=69de5657&is=69dd04d7&hm=8afa58f05c93a1675b1651fc07c9cb89115f5321fc48724a9d65dcc6b5f2778b& 2026-04-13T14:59:50 <.gonzo.> vram is hitting around 13/16gb so it's not hitting a ceiling 2026-04-13T15:04:56 Normal maps make skinning more expensive, so that's something to keep in mind, but I doubt that's relevant here 2026-04-13T15:07:16 So the framerate when not crossing cell boundaries is fine, but the stutter when crossing boundaries goes from small to huge? 2026-04-13T15:07:59 I wonder if it's because we're applying the normal maps with the shader visitor rather than the nif loader. Possibly things aren't being kept cached properly? 2026-04-13T15:10:52 <.gonzo.> this is sophie's performance with all normals and specs on 2026-04-13T15:10:52 <.gonzo.> https://cdn.discordapp.com/attachments/260443767482482688/1493267231487033344/2026-04-13_09-05-28.mkv?ex=69de58fb&is=69dd077b&hm=ca6de24c4c74d9dab7b6d3383326317a40202e9e2eacd85765b9fe077e36fc87& 2026-04-13T15:14:31 <.gonzo.> the 4090 is a fair bit better to be fair, but still it seems a little bit too much worse 2026-04-13T15:17:48 <.gonzo.> also i would say my experience on the 3080 had been pretty similar 2026-04-13T15:18:13 <.gonzo.> i never tried turning normals off on that card, though 2026-04-13T18:02:54 in our testing with CM with HQ textures, the cell change massive performance hit turned out to be lack of mipmaps. 2026-04-13T18:13:52 <.gonzo.> compare the videos. it's the exact same set-up mod-wise 2026-04-13T18:19:56 still worth to check. 2026-04-13T18:20:17 <.gonzo.> is there an ez way to check a billion textures? 2026-04-13T18:29:33 <.hemaris> i know ordenador reports missing mipmaps 2026-04-13T18:35:42 <.hemaris> kartoffels was working on an openmw-specific batch processing tool too https://discord.com/channels/260439894298460160/1452448843403890709/1452448843403890709 2026-04-13T18:38:32 I've done it with python scripts before to validate mipmap normals being normalized for all of my textures, not sure about an existing tool and I have no idea what I did with the script when I was done. 2026-04-13T18:43:58 <.gonzo.> claude the clanker wrote a handy batch script and i'm running through the entire folder tree now with imagemagick 2026-04-13T18:46:00 <.gonzo.> oh i was wrong it was señor gippity 2026-04-13T19:35:22 <.gonzo.> yeah i made sure mipmaps were generated for every file in textures in the entire vfs and it didn't improve the situation 2026-04-13T19:59:58 are the two installs running from the same storage device (ssd/hdd)? 2026-04-13T20:00:02 same filesystem? 2026-04-13T20:01:43 <.gonzo.> yeah it's running off of an SSD 2026-04-13T20:03:44 <.gonzo.> it's a pci gen 3 nvme, so like not the fastest out there, but not slow by any means 2026-04-13T21:40:19 @anyoldname3 did you look at https://gitlab.com/bits-n-bites/buildcache for caching stuff in the CI? 2026-04-13T21:48:11 We got ccache working, and the only slow things are downloading things (which caching won't help with as we need to download the cache), compiling things (which ccache handles), and linking (which will always have unique output). 2026-04-13T21:49:42 Oh, and stuff like zipping up the results of a build, which obviously is unique per build.