A Parallel Algorithm Development Model for the GPU Architecture

Parallel computing has been in use for decades, and throughout many researchers have sought to define a model for algorithm design for such a platform. Valiant developed a model for parallel computing, which was later extended to later include multi-core processors, but it still may not be best suited for the unique GPU architecture. With the current advances in high performance computing, it is easy to see the role that GPUs can play, and even easier to see the need for a model for GPU algorithm development. Here we propose a parallel GPU model which offers both a general design and a fine-grained approach, intended to accommodate nearly any GPU architecture. We show how our model can result in significant increases in performance when algorithms are designed based on its principles.