![]() Man, and that's just at a resolution lower than 1080p in the example one guy was showing: That could be as much as 160 MB per layer (call it maybe 10-14 layers with all the post fx turned on, but maybe 6 with them off. A couple of them also run antialiasing - either MSAA 2x, or FXAA, or both. I made some levers so you can turn off all the various post processing effects, but there are still some cameras that are composited (like six), so that still requires rendertextures. How much RAM do you have? It sounds like some of the post processing chain is insanely heavy in the format it's using for that particular GPU/OS combo when it's above a certain level. Since this persists even when no post-processing effects are on at all, I can tell this is at a very low level. So this is likely why you're seeing an issue with this one game on those drivers: they've got an error in their HDR math, but it's in a part of the HDR range that most games don't use. Because of the needed intensity to make that look right, that means I have to push those higher into the HDR range before clamping them back down after the bloom process. Rather than it being a low-grade scene effect, I'm using it as a high-intensity diffuse effect specifically for engines and for weapons fire. Normally in a terrestrial game with traditional bloom, having those specular highlight bloom slightly is actually a good thing, but I'm using bloom in a very different way. I doubt most other games use such a wide HDR range, but for me it's really useful for being able to not have specular highlights caught by the bloom effect. I do use some quite large HDR emission values at time, and then clamp those back down to the LDR range, and my best guess is that AMD has an error in their math for that on Metal. I realize that probably most games don't have this specific problem for whatever reason, but clearly AI War 2 is getting caught in some sort of strange edge case. Hopefully AMD updates their drivers at some point. When the thing that was causing NaN pixel colors goes away, you'd assume that a camera clear would solve the problem, but instead it stays in some sort of mysterious cache somewhere below the engine. That's some serious low-level driver bugs right there. It looks like NaN pixel color propagation to me, but why it would persist past a camera clear pass (aka from one frame to the next) is particularly mysterious. I suspect that the Radeon drivers, relating to metal, have some bugs with HDR camera support, is all I can figure. ![]() But even without bloom on, the problem happens. It's clearly some sort of screen space error, probably to do with camera compositing, but it's also reacting to bloom for sure. But after literally stripping out every possible post-processing option and the problem persisting on Radeons, I had to add back in OpenGL support just for that one (minority even among Macs - apparently just some desktops from a couple of year period, and no laptops at all). I had actually removed OpenGL support from OSX, because it's so much slower and Metal seemed to be working perfectly. On Radeon only, only on Metal, there's a number of errors in the post-processing pipelne that even turning off all bloom and similar effects does not solve (there are options in the game advanced settings for turning off literally every post-processing full screen effect at this point). ![]() Anyone with an Intel, nVidia, or Silicon GPU can use Metal with AI War 2 just fine. There's an intractable bug with Radeon drivers on OSX that affects Metal only. The TLDR is in the title, but I figure I'd also give the technical explanation in here: ![]() I figured I would go ahead and post this from a recent discussion. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |