Multi-threaded Kernel Offloading to GPGPU Using Hyper-Q on Kepler Architecture