On Green and Energy-Aware GPU Computing for Scientific Applications

Recently, modern graphics processing unit (GPU) has gained the reputation of computational accelerator that can achieve a significant increase in performance by reducing execution time for the different type of scientific application that demand high performance computing. While modern GPUs reduce the execution time of a parallel application as compared to the CPU implementation, but this performance is sometimes achieved at an expense of considerable power and energy consumption. This paper seeks to characterize and explore the impression of high power consumption in a GPU. We examine this notion by reviewing techniques used by researchers to analyze the performance, power, and energy characteristics of GPUs that are utilized for scientific computing. These studies consider applications that run on a traditional CPU setup, and the transformed parallel applications, running on hybrid CPU+GPU environment. These studies indicated that the heterogeneous CPU+GPU environment delivers an energy-aware and sustainable product that is much better than a traditional CPU application.