Scholarlink

This paper presents an algorithm to combine GPU computations for the two-dimensional discrete element method (DEM) with MPI data transfers between computer cluster nodes in distributed-memory systems for one-dimensional domain decomposition. The proposed algorithm redistributes particles between processes, deleting them or adding to the list of particles inside a current subdomain, respectively, in parallel with main GPU calculations to decrease the time loss of calling transferring functions. Strong and weak scaling tests were conducted for the implemented algorithm to show its effectiveness in adding more computational resources to the fixed-size problem and increasing the problem size according to the allocated resources.

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

Multi-GPU Implementation of the Numerical Algorithm Enhancing Discrete Element Modeling Performance

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • V. D. Chepelenkova,
  • D. I. Prokhorov,
  • V. V. Lisitsa

摘要

This paper presents an algorithm to combine GPU computations for the two-dimensional discrete element method (DEM) with MPI data transfers between computer cluster nodes in distributed-memory systems for one-dimensional domain decomposition. The proposed algorithm redistributes particles between processes, deleting them or adding to the list of particles inside a current subdomain, respectively, in parallel with main GPU calculations to decrease the time loss of calling transferring functions. Strong and weak scaling tests were conducted for the implemented algorithm to show its effectiveness in adding more computational resources to the fixed-size problem and increasing the problem size according to the allocated resources.