Scholarlink

The Modular Supercomputer Architecture (MSA) integrates various processing units, with compute modules tailored for specific algorithms, forming a unified heterogeneous system. Modules operate as parallel clustered systems, interconnected via a common or federated network. Optimized resource management allows flexible node selection, aiming for better performance. Hardware heterogeneity poses challenges for developers, who must port and optimize codes for various configurations. This study evaluated the performance of two applications-TSMP (regional Earth-system simulation) and mAIA (fluid dynamics)-across two MSA platforms (DEEP, Wisteria/BDEC-01), considering both hardware and software configurations.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Portability of Multiphysics Applications on Heterogeneous Modular Supercomputers

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Daniel Caviedes-Voullième,
  • Seong-Ryong Koh,
  • Stefan Poll,
  • Estela Suarez,
  • Takashi Arakawa,
  • Kengo Nakajima,
  • Shinji Sumimoto

摘要

The Modular Supercomputer Architecture (MSA) integrates various processing units, with compute modules tailored for specific algorithms, forming a unified heterogeneous system. Modules operate as parallel clustered systems, interconnected via a common or federated network. Optimized resource management allows flexible node selection, aiming for better performance. Hardware heterogeneity poses challenges for developers, who must port and optimize codes for various configurations. This study evaluated the performance of two applications-TSMP (regional Earth-system simulation) and mAIA (fluid dynamics)-across two MSA platforms (DEEP, Wisteria/BDEC-01), considering both hardware and software configurations.