AcHEe: Evaluating approximate computing and heterogeneity for energy efficiency

Abstract

Energy efficiency is lately a major concern for computer engineers, at the levels of both software and hardware. A popular path is the exploitation of heterogeneity and accelerator-based systems, which combine different architectures, each appropriate for specific computational patterns. Approximate computing is another aggressive, yet viable alternative; it minimizes the energy footprint of applications at the expense of output quality.We introduce AcHEe, a set of 12 applications – ranging from real-world applications to kernels – from different domains, exposing a wide range of characteristics, which have been modified to exploit both heterogeneity and approximations. The degree of approximation can be controlled at runtime. The implementation is based on OpenCL and is publicly available.We evaluate these applications on heterogeneous platforms (comprising of CPUs and GPUs) and quantify the isolated and combined effect of heterogeneous and approximate computing. AcHEe can serve as a benchmark suite, to evaluate current and future computational devices from the perspective of heterogeneity and approximations. At the same time, it can serve as a software platform to evaluate and compare different approximation techniques. © 2017