Is it worthing?

I’m going to code a KDTree system with a dynamic occlusion system using the extension nv_occlusion_query. I mean, using a frustum culling, I will only render sectors from KDTree in the frustum, and when render a sector I check via the nvidia extension if the bounding box of the sector is visible on screen, if so I’ll render the sector else no rendering. Do you think that is worthing? I probably have many sectors (about hundreds), I know this is depending mostly on the scenary, but it’ll be mostly indoor scene. Did someone code something like that? How great is it working then?


If you already have the tree and the frustum, then it’s trivially simple to text for sector inclusion using plane/box tests on the CPU. Doing so is likely to be faster and MUCH lower latency than using the pixels-drawn counter.

pixels-drawn really is best if you run the test one frame, but read the results of the test next frame, to make determinations about things like what LOD to draw something in.