Main Scientific and Technological Problems in the Field of Architectural Solutions for Supercomputers

Andrey Molyakov
2020 Computer and Information Science  
In this paper author describes creation of a domestic accelerator processor capable of replacing NVIDIA GPGPU graphics processors for solving scientific and technical problems and other tasks requiring high performance, but which are characterized by good or medium localization of the processed data. Moreover, this paper illustrates creation of a domestic processor or processors for solving the problems of creating information systems for processing big data, as well as tasks of artificial
more » ... ligence (deep learning, graph processing and others). Therefore, these processors are characterized by intensive irregular work with memory (poor and extremely poor localization of data), while requiring high energy efficiency. The author points out the need for a systematic approach, training of young specialists on supporting innovative research.
doi:10.5539/cis.v13n3p89 fatcat:qt6k5wpxi5hytbnifubt4l2vca