I'm reading about efficient Frustum culling algorithms. I found an article about a smart method that first use the Frustum AABB (Axis Aligned Bounding Box) to eliminate most of the scene before check again against the actual frustum representing the camera.
I didn't do any performance test yet, but maybe somebody else did and can answer this question.
Let's say I'm using an Octree to quickly check again the camera bounding box. But I do not want to do a second pass again the camera planes. If I simply render everything so far, isn't that actually faster than do the second pass?
Assume object geometry is loaded into VRAM using hardware vertex buffers.