PDA

View Full Version : backface culling performance



mordrax
09-16-2005, 12:32 AM
i'm doing a little performance test to see how good backface culling is for an assignment.
I'm loading in a OFF format file, basically specifying 26k verticies and 46k faces, all polygons. It's a model of a harley davidson :)

Now, on my GeForce 6600GT PCIe, the speed hardly changes from backface cull to render everything. With culling on, it's about 150fps, with it off, it's about 148fps...

i'm doing a glfinish on every frame if that helps, but i can't for the life of me see why i'm getting no improvements with the culling. i've tested it, by moving into the picture n out, there's nothing wrong with the culling that i can see, it's working properly...

confused...

Joe.

mordrax
09-16-2005, 12:35 AM
couldn't find an edit button

i meant, all triangles, not all polygons :p

memfr0b
09-16-2005, 01:33 AM
The benefit from backface culling very much depends on where your application is limited. With backface culling you save rasterizing, texturing, z/alpha/stencil testing fragments of triangles that aren't visible anyway. So unless you application is fillrate limited you won't get any benefit culling, because it doesn't change the number of vertices sent to the card, or the amount of cpu work necessary.

You might get better results by reducing the work in the other areas, for example by using triangle strips instead of triangles, or using VBOs / display lists for vertex storage. Or artificially increasing the workload per fragment through blending, texturing or using a complex pixel shader.

Bottom line: benefit from culling is highly application dependant.

PS: the edit button is the little pen-and-paper icon on the top of every post.