Scalable custom computers with less influence of a memory bottleneck for next-generation massively-parallel fluid dynamics computation

2014 Fiscal Year Final Research Report

• Project/Area Number: 23300012
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Computer system/Network
• Research Institution: Tohoku University
• Principal Investigator: SANO Kentaro 東北大学, 大学院・情報科学研究科, 准教授 (00323048)
• Co-Investigator(Kenkyū-buntansha): SASAKI Daisuke 金沢工業大学, 工学部, 講師 (60507903) ; YAMAMOTO Satoru 東北大学, 大学院情報科学研究科, 教授 (90192799)
• Co-Investigator(Renkei-kenkyūsha): NAKAHASHI Kazuhiro 宇宙研究開発機構, 理事 (00207854)
• Project Period (FY): 2011-04-01 – 2015-03-31
• Keywords: 専用ハードウェアアクセラレータ / 数値流体力学 / ビルディングキューブ法 / リコンフィギャラブル計算 / FPGA / 計算機アーキテクチャ / 高性能計算 / 並列計算

Outline of Final Research Achievements

We have researched hardware accelerator architectures for massively-parallel fluid dynamics simulation based on the building-cube method (BCM). The accelerators achieve high performance and scalability with such a custom hardware structure for stream-computing that is less influenced by a memory bottleneck. After developing an infrastructure system consisting of a tightly-coupled FPGA cluster, an FPGA-based system-on-chip, their software stacks, and a high-level synthesis hardware compiler, we evaluated a prototype system implemented with them to show that the proposed hardware structure and data-exchange mechanism are suitable for highly-efficient and high-performance fluid dynamics computation based on BCM.

Free Research Field

総合領域