Links to the performance metrix (CPU<-> GPU communication, queue performance, core usage etc ) and response times ( min, max, average, standard deviation of services/programs) quaranteed / expected /promised by AMD for the OpenCL Implementation more than welcome !
thanks in advance!
You can refer to OpenCL Programming Guide sec 4.5.
You can get buffer transfer rates obtained on a specific system system in different cases. Also refer to the enchmark samples that comes with SDK to see how much it improves for you.