quote:
How fast is OpenGL''s view volume clipping?

The clipping itself is extremely fast. Changing the clipping planes however, is an expensive state change. It will vary with hardware and drivers. Basically, it''s a matrix update (projection matrix), along with a clipping plane extraction. It shouldn''t hurt, if you do it two or three times per frame, but not for every single object.

But why would you want to do that ? I guess you want to use a different far plane to cull away small objects ? Not a good idea. If you are after better performance, then you''ll need to do the high level culling on the CPU, using some form of hierarchical object structure.

The bottleneck of most modern engines is not the rendering speed of the 3D card, but the transfer speed of the AGP bus. Leaving the culling to the 3D card won''t help in that situation, as your data is still transfered over the bus, just to be discarded by the hardware later on. Pretty much a big waste of resources. If you cull your object on the CPU, then it''s data isn''t transfered anymore. This will speed your engine up.

/ Yann

The problem is exactly the same with VAR/VAO. High level culling should (must) be done on the CPU. That''s what structures such as octtrees are for. Approximating highlevel culling through lowlevel culling on the GPU is a waste of resources, in terms of transfer, transform pipeline saturation (a vertex is still processed by the GPU, before being culled), and VRAM/AGP memory usage.