>> Saturday, May 4, 2013
In OpenCL, transferring data between the CPU and GPU can take a significant amount of time. This is a problem for all GPGPU applications, and to solve it, AMD has developed the Heterogeneous Uniform Memory Access, or hUMA. Ars Technica has a great article on the subject here.
It looks wonderful to me, particularly the cache coherence. With hUMA, CPU-GPU data transfer will only consist of sending pointers back and forth. And if this will be available for the upcoming PS4, that will be even better.