ML-Gov

Abstract

Modern heterogeneous CPU-GPU based mobile architectures that execute intensive mobile games and other graphics applications use software governors to achieve high performance with energy-efficiency. For dynamic and diverse gaming workloads on heterogeneous platforms, existing governors typically utilize statistical or heuristic models assuming linear relationships for a small set of mobile games, resulting in high prediction errors. To overcome these limitations, we propose ML-Gov: a machine learning enhanced integrated CPU-GPU governor that builds tree-based piecewise linear models offline, and deploys these models for online estimation into an integrated CPU-GPU Dynamic Voltage Frequency Scaling (DVFS) governor. Our experiments on a test set of 20 mobile games exhibiting diverse characteristics show that our governor achieved significant energy efficiency gains of over 10% improvements on average in energy-per-frame with a surprising-but-modest 3% improvement in Frames-per-Second (FPS) performance, compared to a typical state-of-the-art governor that employs simple linear regression models.