Analysis on Heterogeneous Computing

摘要

Abstract In the Internet industry, with the popularization of informatization and the rapid increase in data volume, people have new requirements for storage space. At the same time, computer applications such as artificial intelligence and big data have rapidly increased demand for computing power and diversified application scenarios. Heterogeneous computing has become the focus of research. This article introduces the choice of architecture for heterogeneous computing systems and programming languages for heterogeneous computing. Some typical technologies of heterogeneous computing are illustrated, including data communication and access, task division and mapping between processors. However, this also brings difficulties. The challenges facing hybrid parallel computing, such as programming difficulties, poor portability of the algorithm, complex data access, unbalanced resource load. Studies have shown that there are many ways to improve the status quo and solve problems, including the development of a unified programming method, a good programming model and the integration of storage and computing, intelligent task allocation, as well as the development of better packaging technologies. Finally, the application prospects and broad market prospects of heterogeneous computing systems are prospected. In the next ten years, due to the various advantages of heterogeneous computing systems, innovation in more fields will be stimulated and heterogeneous computing systems will shine in the AI artificial intelligence fields such as smart self-service equipment, smart robots, and smart driving cars. Moreover, this emerging technology will bring new industries and new jobs, thereby driving economic prosperity and social development and even benefiting the entire human society.