Unified memory architecture! Should be some cool hybrid cpu/gpu algorithm possibilities. Currently the overhead of copying stuff between the two memories in CUD

Unified memory architecture! Should be some cool hybrid cpu/gpu algorithm possibilities. Currently the overhead of copying stuff between the two memories in CUDA prevents any fine-grained cooperation. I wonder how much gpgpu ability is exposed.