Energy-efficient visualization using MegaMol

Abstract

Because of rising ecological awareness and higher cost of living, power efficiency is becoming an increasingly deciding factor in most technology, including computers. However, a recent trend with GPUs, already one of the most power-hungry computer components, is to allow higher maximum power draw than ever before. While there is research on the power efficiency of GPGPU applications, even going as far as modelling power consumption at the instruction level, there is next to none specifically on scientific visualization and how factors like dataset size or rendering approach affect power consumption and efficiency. Additionally, most research focuses solely on GPUs by manufacturer Nvidia, although other manufacturers exist and also strive to improve computing power and efficiency. This work analyses the effects of various factors on power consumption and efficiency of different scientific rendering approaches (sphere splatting and raycast volume rendering) across a selection of high-end consumer GPUs by the three common manufacturers of (AMD, Intel and Nvidia). Power consumption is measured with the integrated power sensors integrated into most modern GPUs and an external hardware measuring setup. MegaMol, a scientific visualization platform with automation capabilities, is the framework for the benchmarks.