Traditional Architecture Artificial Intelligence Chip Technology

Abstract

In this paper, we take three kinds of traditional artificial intelligence (AI) chips as the research object: GPU, FPGA and ASIC, introducing the architecture characteristics and advantages of some famous products which have large influences on the development of this technology, such as NVIDIA V100, Cyclone V, Stratix 10 and so on. The traditional AI chips, including GPU chips, FPGA chips, and Low Power AI ASIC, are analyzed in this article. The main focus is on the cutting-edge technologies and designs in artificial intelligence based on these traditional AI chips in recent 10 years, especially the applications on neural networks and model trainings, such as Kubernets+Docker Container, S2N2 (an FPGA accelerator), FNSim (a simulator of flash), which have become hot areas of scientific research for several years and will be brought on stream widely by scientific and technical corporations in the industry.