Scholarlink

This report discusses a model to improve the performance of Global Climate Models (GCM) and Air Quality Models (AQM). Both models share an implementation of the Gear solver for the aqueous chemistry transport model (CTM). A thread level version of CMAQ has been developed by HiPERiSM over the last decade and this has shown a reduction in wall clock time by factors of 1.2–1.5 depending on which of the three solvers the CTM uses. This report is focused on the Gear solver in an OpenMP thread enabled HiPERiSM (FSparse) version when part of the CTM is offloaded to a graphical processing unit (GPU). A model of the compute intensive part of the Gear solver shows that the most expensive part of the calculation, when offloaded to a GPU device, is enhanced by a factor of 10 compared to the same calculation on a CPU host.

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

Accelerating Performance in Global Climate and Air Quality Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • George Delic

摘要

This report discusses a model to improve the performance of Global Climate Models (GCM) and Air Quality Models (AQM). Both models share an implementation of the Gear solver for the aqueous chemistry transport model (CTM). A thread level version of CMAQ has been developed by HiPERiSM over the last decade and this has shown a reduction in wall clock time by factors of 1.2–1.5 depending on which of the three solvers the CTM uses. This report is focused on the Gear solver in an OpenMP thread enabled HiPERiSM (FSparse) version when part of the CTM is offloaded to a graphical processing unit (GPU). A model of the compute intensive part of the Gear solver shows that the most expensive part of the calculation, when offloaded to a GPU device, is enhanced by a factor of 10 compared to the same calculation on a CPU host.