are you sure about your post about how the occlusion works? I only know occlusion culling done on the gpu. This aside rendering something on the cpu, writing a software renderer for it etc. just seems to be utterly mad considering you can just render low res images on the gpu as well using another render target or stuff (not familiar with the dx terminology but in opengl that is straight forward and mutipass renderings for postfx, shadow maps etc aren't that big of a deal). I don't know about the implementation in reflex but I think the only benefit from turning off sth like occlusion culling would be a worse but more stable frame rate because there might be parts in a level where most of the geometry is culled vs parts where less is culled thus leading to a very different rendering load etc.