It's unclear to me that advantage of using UAVs over global memory, does anyone know what this might be or where I can find this information, the CAL docs are, again, very thin pickings.
ryta, Global buffer is 128bit aligned addresses, UAV's are byte aligned and on 5XXX series of cards you can have up to 9 UAV's per kernel. Also through UAV's you can do byte addressable writes with the UAV arena and also atomic operations. None of these can be done on the global buffer path.