A64FX: 52 Core Processor Designed for the Supercomputer Fugak

• Published in: Proceedings of Technical Program of International Symposium on VLSI Design, Automation and Test p. 1
• Main Author: Tabata, Takekazu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.04.2022
• Subjects: Artificial intelligence; Large-scale systems; Parallel processing; Power demand; Reliability engineering; Supercomputers; Very large scale integration
• ISSN: 2472-9124
• DOI: 10.1109/VLSI-DAT54769.2022.9768080

Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. In supercomputing, high-performance computing (HPC) applications require large amounts of parallel processing capability and high memory bandwidth. Recently, artificial intelligence (AI) applications, which have characteristics similar to HPC applications, have played an important role in supercomputers. The supercomputer Fugaku has a large number of general-purpose processors for high performance, high-density implementation as well as low power consumption to realize a massive parallel system for multiple applications targeting HPC and AI. Furthermore, in terms of availability, reliability in large-scale systems such as Fugaku is important. Fugaku is a supercomputer comprising >7.6 million cores and has remained No.1 in the Top 500 for three consecutive terms because it achieved 442 Petaflops at 29.9 Mega Wat in June 2020. In this session, for large-scale systems, we describe how CPUs in Fugaku solve challenges of high performance, low power consumption, and reliability.