Programming models for “Fugaku” and beyond
Thursday, July 1, 2021 1:55 PM to 2:15 PM · 20 min. (Africa/Abidjan)
Stream#1
Exascale Systems
Information
Contributors:
Abstract:
Fugaku is the Japanese flagship supercomputer, ultra-scale “general-purpose” manycore system with 150K nodes using A64FX manycore processor per a node. A64FX has 48 cores with Arm SVE SIMD instructions. To get high performance of Fugaku, the programming models have to address how to exploit parallelism and performance of SIMD and manycore of A64FX and huge number of nodes in the system. In this talk, the overview and performance of Fugaku and A64FX processor will be presented, followed by the system software of Fugaku and the challenges for programming models for Fugaku. Some perspectives on programming beyond “Fugaku” are also addressed.
Abstract:
Fugaku is the Japanese flagship supercomputer, ultra-scale “general-purpose” manycore system with 150K nodes using A64FX manycore processor per a node. A64FX has 48 cores with Arm SVE SIMD instructions. To get high performance of Fugaku, the programming models have to address how to exploit parallelism and performance of SIMD and manycore of A64FX and huge number of nodes in the system. In this talk, the overview and performance of Fugaku and A64FX processor will be presented, followed by the system software of Fugaku and the challenges for programming models for Fugaku. Some perspectives on programming beyond “Fugaku” are also addressed.
