I had a moment and thought about this issue some more and this could be a little out there but it’s obvious that the issue lies with Lightmass. So let’s assume since Lightmass doesn’t need the real-time reliability and thus passes off all the calculations or at least the majority off to the CPU, and does it’s pass before compiling the final asset used in the Deferred Lighting Pass, I’m guessing there is a possibility NVidia maybe implemented some form of pre-visualization pass or pre-culling that maybe even though the final rendering pass uses interpolation (linear, anisotropic, etc.) maybe the pre-culling works like a point filter does so where there is no interpolation and it could be looking at the results on a per-pixel basis and decides maybe in a simple if statement if it needs to render a specific area and if a given set of rules aren’t met then it ignores it. It’s pretty wild speculation but who knows, hopefully Tim can chime in with some feedback.