Architecture agnostic energy model for GPU-based design

Since 3D graphic applications have reached mobile devices and known a tremendous expansion, graphic processors' power consumption has become a major challenge in mobile SoC design. We propose a high-level hardware agnostic model endorsed by GPU energy consumption measurements under specific 3D graphic applications running. The modeling method comprises of two phases where the first phase is GPU hardware dependent, whereas the second phase takes graphics API trace information only per a target application. Our model shows an average relative error as small as -5.43% in a run-time of 2 seconds for the well-known benchmark glmark2.