项目名称: 面向浅水波大气动力方程求解器的可重构计算方法研究
项目编号: No.61702297
项目类型: 青年科学基金项目
立项/批准年度: 2018
项目学科: 自动化技术、计算机技术
项目作者: 甘霖
作者单位: 清华大学
项目金额: 8万元
中文摘要: 基于高性能计算系统的大气模式数值模拟方法是目前应对重大气候变化问题的重要手段之一,也是国家战略层面的重大需求以及高性能计算技术研究的一个重要研究方向。对于传统高性能计算平台来说,由于受限于内存带宽等因素,在大气模式模拟中往往很难取得理想的计算效率,高性能计算机强大的计算能力不能得到充分发挥。近年来,基于FPGA的可重构计算方法,因FPGA硬件资源可编程特性以及较低的工作频率,逐渐成为实现高性能、低能耗数值模拟研究的一个极具潜力的平台。针对地球系统模式中大气模式动力框架模拟对高性能平台计算效率与能耗效率越来越高的需求,本项目基于FPGA可重构计算平台研究优化方法,从存储、精度等角度出发设计高性能、低能耗的可重构方法,取得在计算性能上至少一个数量级的提升以及在能耗方面至少一个数量级的降低。
中文关键词: 可重构计算;高性能计算;大气模式动力框架;异构计算;FPGA
英文摘要: Global atmospheric modeling is an essential method to study the climate issues, and a popular topic for high performance computing (HPC) research. However, due to the restrictions such as low memory performance, traditional HPC platforms such as CPU, GPU, and MIC are generally facing difficulties in achieving good computing efficiency for atmospheric modeling. On the other hand, reconfigurable system such as FPGA has been playing promising roles in many important applications to have achieved good result in both performance and power efficiency. In this project, we plan to design a novel method for solving the global shallow water equations algorithm based on reconfigurable devices. Through studying reconfigurable optimizations such as memory and precision, we will deploy a hardware workflow to process the numerical algorithm in a better performance and power efficiency.
英文关键词: Reconfigurable Computing ;High Performance Computing;Atmospheric Dynamics;Heterogeneous Platforms;FPGA