嵌入式圖形處理器之繪圖程式高階耗能模型

Abstract

嵌入式圖型處理器可加速行動裝置上的繪圖程式之繪圖處理，但同時也需消耗可觀的耗電[1]。由電池驅動的行動裝置之耗電量無疑是非常重要的。為了能預估繪圖程式的耗電量，過去研究會去收集GPU硬體元件之計數器的數值來作為耗能模型之參數並預測耗能。但這些參數無法協助繪圖工程師來撰寫較省電之繪圖程式。為了讓繪圖工程師了解繪圖品質與耗能的關係，此篇論文建立了一個只需要繪圖參數，不需要硬體計數器即可預測繪圖程是耗能的耗能模型，以協助繪圖工程師在繪圖品質與耗能中取得一個平衡點。此耗能模型的預估值與實際量測的耗能的誤差為7.7%。

Embedded graphic processing unit (GPU) accelerates a real-time rendering process of a graphics application on mobile devices, however, at the cost of consuming a considerable portion of the system energy [1] which is one of the most critical design issues for battery-operated devices. To estimate the power consumption of a graphics application, conventional approaches collect run-time hardware activities of a GPU, and derive the power consumption of the graphics application based on hardware counters. Unfortunately, these hardware counters and power consumption information are difficult to evaluate from a programmer's point of view. In order to provide graphics programmers a firm notion of how performance and quality relate to energy cost, a high-level power model to assist programmers to balance performance, quality, and energy budget is proposed in this study. The high-level power model only requires high-level graphics data to estimate the power consumption of a graphics program. The error rate of the model is around 7.7%.