Exploiting remote GPGPU in mobile devices

Abstract

Smart mobile devices including smart phones and tablets have become one of the most popular devices in the personal computing environment. One of the major characteristics of mobile applications is that the applications in the field of entertainment like games and augmented reality require a great deal of computations. In order to deal with this, smart mobile devices began to be loaded with application processors equipped with high performance GPUs. In this study, the feasibility of having computation intensive mobile applications to use the GPU resource of another GPGPU-enabled device in the same space for their computation tasks was verified. If benefits can be obtained in terms of the performance and energy consumption, by having the high performance GPU of a remote device to perform the complex computations that are currently performed on a local device CPU, such an approach can be used as an essential technology for the distributed computing among mobile devices. In order to verify this, we not only implemented the game 'Reversi' using the Monte Carlo Tree Search (MCTS) algorithm but also implemented a remote GPU support framework to Android platform so that it supports task offloading to GPGPU-enabled remote mobile devices. The Reversi game offloads computationally heavy parts of the MCTS to a remote GPU through our remote GPU support framework. We compared its performance and energy consumption with the case where the MCTS was completely performed on a local CPU. The results showed that (1) the game winning rate dramatically increases and (2) the overall energy consumption greatly decreases when the remote GPU was used.