With today''s graphics cards, is it faster to do a frustrum cull for every unit, or just send them to the video card and let the video card deal with it?

Hi

I think the best way is to calculate a set of possible visible objects (maybe your Scenegraph is able to do that...i.e. portal engines etc). For this objects you should do a Frustum Culling, sort the remaining objects by texture (shader ?) and render them.

This should be the fastest way to render things. If your complete scene is quite lowpoly (i.e. Quake 2 Map + md2 Models) you could just sort them by texture and render all of them, todays graphics cards should render that with 150+ fps (maybe even 300+).

With HW TnL you should never do culling on a per triangle-base, this would slow things down. The best should be a per object or per render call base. This just relies on your texture count/shader complexity.

Bye,
Oesi