Array of structures (interleaved vertex data) is the preferred choice if the vertex data is well packed (i.e. all components are used) and all components have about the same update rate. However this isn't the case in Prism since tex2 is usually unused. This is an additional 25% saving in bandwidth and memory we missed. I believe it is better (certainly the case for VBO) to separate position data and tex data into separate buffers. With this change we can explore using a smaller data type for tex data hence further reduce the bandwidth and memory requirement.

(Filed on behalf of Chien)