Scholarlink

In this paper we investigate the effect of power capping on the runtime and energy-to-solution for two benchmarks and seven HPC codes. We also study the impact of computational load balance on energy efficiency and we can show that power capping and improving computational load balance are independent paths that lead to increased energy efficiency in HPC centers. For some applications, we observe a minimum in energy-to-solution for a certain power cap. We estimate up to which efficiency of the computing center running compute nodes at lower power caps actually saves energy. In order to estimate how much energy can be saved by optimizing computational load balance, we present a statistical approach that uses the distribution of load balance obtained from performance analyses of many codes.

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

Impact of Computational Load Balance and Power Capping on Energy Efficiency in HPC Centers

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Martin Rose,
  • Jose Gracia,
  • Christian Simmendinger,
  • Andreas Ruopp,
  • Ramil Nabiev,
  • Christoph Niethammer

摘要

In this paper we investigate the effect of power capping on the runtime and energy-to-solution for two benchmarks and seven HPC codes. We also study the impact of computational load balance on energy efficiency and we can show that power capping and improving computational load balance are independent paths that lead to increased energy efficiency in HPC centers. For some applications, we observe a minimum in energy-to-solution for a certain power cap. We estimate up to which efficiency of the computing center running compute nodes at lower power caps actually saves energy. In order to estimate how much energy can be saved by optimizing computational load balance, we present a statistical approach that uses the distribution of load balance obtained from performance analyses of many codes.