CAJ | 학술논문

随着GPU的发展,其计算能力和访存带宽都超过了CPU,在GPU上进行通用计算也变得越来越流行,这样就构成了CPU-GPGPU的新型异构体系结构.虽然这种新型体系结构表现出了强大的性能优势并受到了学术界和产业界的广泛关注,但如何更好地在这种结构上高效地编写和运行程序仍然存在很大的挑战.本文综述了针对这一体系结构现有的可编程性技术、可靠性技术和低功耗技术,并结合这些技术展望了CPU-GPGPU这种异构系统的发展趋势.
수착GPU적발전,기계산능력화방존대관도초과료CPU,재GPU상진행통용계산야변득월래월류행,저양취구성료CPU-GPGPU적신형이구체계결구.수연저충신형체계결구표현출료강대적성능우세병수도료학술계화산업계적엄범관주,단여하경호지재저충결구상고효지편사화운행정서잉연존재흔대적도전.본문종술료침대저일체계결구현유적가편정성기술、가고성기술화저공모기술,병결합저사기술전망료CPU-GPGPU저충이구계통적발전추세.
GPU has better performance than CPU in both computing ability and memory bandwidth with the fast development of GPU technology. Therefore general computing on GPU has become increasingly popular, bringing forth an emerging CPU-GPGPU heterogeneous architecture. Although the new architecture demonstrates high performance and are currently being the highlight in academe and industry, how to write and execute programs on it efficiently still remains a big challenge. This paper summarizes the techniques of programmability, reliability and low power for GPU, and discusses the development trend of the CPU-GPGPU heterogeneous architecture.