Scaling applications on cloud using GPGPU- trends and techniques

Abstract

This survey presents an elaborative study of different GPU accelerated cloud frameworks. Although, a school of thought still advocated CPU over GPU computation stating limited on chip memory of GPU, several models have been proposed which have proved GPU can outperform CPU based systems if properly optimized. The authors have discussed the contexts which serve beneficial for such GPU based systems, and have presented examples through literature survey where GPU based computations have outperformed CPU despite limited on-chip memory. The discussion also includes application areas which can be enhanced using CPU-GPU hybrid system and remarks on how to achieve this.