Why we get worse performance when using private mode buffer for vertex/index buffer compared to using shared mode?
rencently, with the new gpu memory profiling tool, we find that our game engine create all gpu resources with shared storage mode on iOS, which is at odd with the recommendation that persistent resouces should be private. So I wrote a simple program to investigate the performance boost we can have. However the result is surprising: shared stoarge mode make vertex/index buffer access faster than using private storage mode. For texture the performance is pretty close. The attachment are a detailed report and the test program.
