Co-Design for A64FX Manycore Processor and ”Fugaku”

Sequences of Sparse Matrix-Vector Multiplication on Fugaku’s A64FX processors

“…A major feature of the recent ARM-based processors, including the A64FX, is the Scalable Vector Extension (SVE) that enables support for SIMD operations with per-lane prediction, which allows for efficient vectorization [10]. The vector length can be specified as a multiple of 128 bits.…”

Co-design in the Exascale Computing Project

“…While the remainder of this overview will discuss how this goal was realized within the U.S., and in particular DOE, similar efforts were of course underway in other countries. Perhaps the most notable among these is the Flagship 2020 project in Japan, which culminated in the development of Fugaku, as described in Sato et al (2020).…”

Comparing the behavior of OpenMP Implementations with various Applications on two different Fujitsu A64FX platforms

“…In this (short) paper, we analyze the OpenMP [7] shared-memory/parallel programming model, from how it scales on the A64FX -and its variants on Ookami and Fugaku, see Section 2 -to performance across different compiler toolchains. 2 The A64FX processor [6,10] is the processor used in the Fugaku supercomputer, enabled by the Japanese FLAGSHIP 2020 project as a co-design between RIKEN and Fujitsu. Currently, Fugaku is ranked number 1 on both the Top500 and HPCG lists.…”

“…Currently, Fugaku is ranked number 1 on both the Top500 and HPCG lists. A64FX is a generalpurpose processor based on the Armv8.2-A specification [10] and has 48 compute cores, divided into four core memory groups (CMGs) with 12 cores, and 2 − 4 cores dedicated to OS communications.…”