I normally work with the compositor at 256x256 with OpenCL ticked… I have A Corei5 3570k and a HD7970 GPU.
Today i found that if i un-ticked the OpenCL it shaved a few seconds off the calculation time. I was surprised.
Its either a really badly optimized OpenCL on the HD7970, OR it was using the gpu inside the Core-i5 to run the OpenCL… Which is of course a pointless waste of die space by intel because its no faster than the cpu anyway, and everyone has a discreet video card.
Anyway ive always wondered which compute device the compositor uses for its OpenCL.