GPU/CPU协同粗粒度并行计算及在城市区域震害模拟中的应用

GPU/CPU Cooperative Coarse-grained Parallel Computing and Its Application to Regional Seismic Damage Predicction

采用精细结构模型和动力时程分析以提高城市区域建筑震害预测精度已经成为一重要研究方向,而传统的CPU计算平台成本过高,难以推广。本文提出采用基于GPU/CPU协同粗粒度并行计算的方法来实现城市区域建筑震害的高效精细化动力时程计算,可以显著提高效率并降低成本。简述了所采用的程序架构、计算模型、参数选取,对并行计算的效率进行了详细的讨论,并通过一个中等大小城市的案例展示了该方法的优势。

Refined model and dynamic time-history analysis have been deemed as a promising way to improve the accuracy of seismic damage prediction in urban city. However the high cost of tra- ditional CPU computing platform hinders the extensive use of this method. In order to solve this problem, the GPU/CPU cooperative coarse-grained parallel computing approach, featured in high efficiency and low cost, is adopted to realize the regional seismic damage prediction with dynamic time-history analysis and refined structural model. This paper introduces the program architec- ture, calculation model and the parameters selection of the proposed method. The efficiency is discussed in detail and the method is implemented to a mid-size city, by which the advantages of this method are demonstrated.