On May 15, 2012, NVIDIA announced several innovative technologies aimed at improving the performance and energy efficiency of the Kepler GPUs, along with opening up a new class of applications and algorithms – creating a larger ecosystem of developers and applications for the GPU landscape.
Two of these innovations in particular caught our attention.
The first is Hyper-Q—a technology that enables multiple CPU cores to simultaneously use the CUDA architecture cores on a single Kepler GPU. The idea here is to dramatically increases GPU utilization while slashing CPU idle times and advancing programmability. Hyper-Q is enabled in CUDA 5, and we can expect to see it used heavily with standard MPI codes. Hyper-Q offers great promise for dealing with concurrency issues as we scale to much larger systems.