Control theory-based DVS for interactive 3D games

We propose a control theory-based dynamic voltage scaling (DVS) algorithm for interactive 3D game applications running on batterypowered portable devices. Using this scheme, we periodically adjust the game workload prediction based on the feedback from recent prediction errors. Although such control-theoretic feedback mechanisms have been widely applied to predict the workload of video decoding applications, they heavily rely on estimating the queue lengths of video frame buffers. Given the interactive nature of games - where game frames cannot be buffered - the controltheoretic DVS schemes for video applications can no longer be applied. Our main contribution is to suitably adapt these schemes for interactive games. Compared to history-based workload prediction schemes - where the workload of a game frame is predicted by averaging the workload of the previously-rendered frames - our proposed scheme yields significant improvement on different platforms (e.g. a laptop and a PDA) both in terms of energy savings as well as output quality.