Just wanted to give a heads-up that other Cycles commits in master have made the D13533 patch no longer able to be applied. Any chance someone can take a look at refreshing the diff to work with the latest master? If not, I can try my hand at updating it, but I’m probably the second-worst person to do it, just after the local beauty pageant winner.
EDIT: Got a working fix - posting as a comment to the diff itself.
1 Like
Is MNEE only for CUDA rendering Device? I downloaded the 3.2 alpha build today and tried on optix but it seems to crash every time. If it should be working on optix I’ll submit a bug report.
It works for me with OptiX, but the first-time kernel loading is very very long, maybe something is wrong with it
@Eary and @Bobo_The_Imp Both myself and others have experiencing issues with MNEE on OptiX unless we are using very recent GPU drivers. Try updating your GPU drivers to the latest version and see if it helps.
I’m on a fully up to date 3090. I also tried the patch from back in the day before it hit master. It was the same on then as well. Unless I switched to CUDA it would crash, but I assumed it was just because it was a patch.
Hi.
Is the method expected to be heavy/slow when you use high vertex count meshes, instead of low vertex count meshes and Normals Auto Smooth?
I know most people here have modern GPUs, but my GTX 960 is suffering in those cases:
Cycles_Glass_Render_Time.blend
I’ve been looking through papers recently for improvements in this area, and came across this paper: Specular Manifold Bisection Sampling for Caustics Rendering
The results appear to be an amazing improvement over SMS, with much better visuals at the same time taken - could be a good addition instead of normal SMS once this eventually gets more work done on it.
3 Likes