In order to reduce the vertex bandwidth in our title, I check to see if the number of indices in a vertex buffer is < 256. If so, I create a VBO with byte indices.
When it comes to rendering, I check the number of indices, if < 256 I use:
otherwise I use:
This reduction of bandwidth actually reduces my framerate by about 5%!
Are the byte elements getting converted to shorts in the driver or something?