Ok, lets say I am making a cliff / rock wall from lots of instances (lets say 8 different models instanced a 1000 or so times). These models are quite spherical and are rotated in every which direction, and are intersecting all over the place. Is there much of a rendering performance cost associated with doing things this way? as there will be a whole lot of geometry on the backfaces etc that never get fully rendered. (lets say 90% of the total polycount isn’t visible to the camera). Is it a bad idea to do things this way?
will the geo that isn’t seen by the camera even have a cost to rendering that is worth worrying about? (only faces will be non visible to camera, not whole objects, so i guess they wont get occluded)
is there a link to any docs that cover this? i read the deferred rendering overview but it didn’t really answer my questions.
Any info will be greatly appreciated