Trying to understand if Unreal 5 uses optimizations under the hood (on NVIDIA there’s VRWorks MVR, AMD has something similar I think) to make cubemaps and other multi cam drawing more efficient. Trying to understand if an algorithm requiring multiple cameras is too expensive.
From my experiments, enabling a cubemap that renders every frame creates a perf hit from 116fps to ~15fps. That makes sense as it’s 7 cams instead of 1. However, reducing the target size in each face to 32 from 1024 doesn’t produce significantly better rendering performance, as you might expect if the render was shading bottlenecked. This indicates the hit is in the view setup per face.
Anyone know if there is a way to best amoritize setup costs here?