Accelerator architectures: A ten-year retrospective

Abstract

This article is a ten-year retrospective of the rise of the accelerators since the authors coedited a 2008 IEEE MICRO special issue on Accelerator Architectures. It identifies the most prominent applications using the accelerators to date: high-performance computing, crypto currencies, and machine learning. For the two most popular types of accelerators, GPUs and FPGAs, the article gives a concise overview of the important trends in their compute throughput, memory bandwidth, and system interconnect. The article also articulates the importance of education for growing the adoption of accelerators. It concludes by identifying emerging types of accelerators and making a few predictions for the coming decade.