thomas wrote:Ok, here's one: Multiple camera's per scene.
The camera's would be rendered simultaneously sharing light paths and memory. Useful for walkthrough animations, archviz with multiple pov's, 3d stereo, etc.
While this sounds good on paper, I suspect the speed-up is minimal and the added complexity would be great (not to mention memory usage for larger resolutions). The problem is that generating the light path is a relatively small part of the whole pie. The majority of intersection tests (which in total stands for 60-80% of the cpu time) comes from connecting the eye and the light paths, and since the eye paths will be different there's little to share here.
For animation and such I suspect photon mapping or something similar would be a better approach.