try staying away from polygons. hardware likes triangles and everything else would have to first be turned into triangles. also polygons have the annoying habit that unlike triangles they can easily be screwed up and arent on a simple plane anymore. so i guess theres a reason why even nvidia exsamples use a triangle strips instead of a quad (though i cant imagine the difference will be that huge).

i store my data twice once the world data in polygon classes for collision detection and the static render data in system memory
if the gfx hardware supports VBOs i store it there

for models you only need to store the transformation matrices in system memory

@Trienco: i don t use triangles for collision detection since it would be too computitional expensive

Ah by polygon I meant triangle, but they are triangles stored in a Polygon class. ^^;;