Im working on a ROAM implementation and just finished backface culling. I dont see any improvement though and im wondering if there''s a limit to how many triangles its feasable to backface-cull. I have all normals precalculated so only the test is run each frame. The scene so far consist of approximately 3500 triangles. Can the lack of performance-improvement be because the test takes too long to perform ? or is 3500 triangles not too much to to test for backfacing ? Thanks in advance .

If your choke point isn''t geometry (or maybe fill rate) then you won''t see any improvment by backface culling. Try cranking up the number of triangles more. (A LOT more if you have a recent video card). Then it should make a difference.