Everyone is now using OpenCl
Everyone is now using OpenCl to accelerate more computing on the GPU, and eventually the CPU cores on future HSA compliant versions of the APU will actually directly dispatch floating point calculations to the GPU’s floating point units without having to go through any software abstraction layers. I can see future APU’s CPU cores and their GPU accelerators becoming so integrated that it will be hard to tell them apart, and that includes the chaining of 32 bit floating points units to make double and quad or more precision calculations possible in the hardware, so it’s going to be like having variable width FP units in blocks of 32 bits up to maybe 1024, or 2048 if that is ever needed. I can see even more of any computing software moving as much computations onto the GPU as possible, and any CPU bound hardware not HSA compliant as being at a definite disadvantage.