大型非線性結構在進行分析時,常因為運算量太過龐大,而導致電腦運算時間過長。因此在分析上,折衷採用分析運算量較少但精度較低的數值模型。為了要解決運算量的問題,各式各樣的高速運算方法不斷的運用在科學運算中。GPGPU的超多核心與低成本的特性,使得GPGPU有潛力成為進行大型非線性結構分析的另一個工具。 隨著科技發展越來越進步,繪圖處理器 (Graphics Processing Unit,簡稱GPU)具有「超多」核心(many-core)的特性,其運算能力隨之變得非常強大。近年來,GPU的超多核心的特性更進一步地被發展為適用於一般運算用途的處理器,而此技術被稱為GPGPU(General-Purpose computing on Graphics Processing Unit)。本研究重新檢視OpenSees(Open System for Earthquake Engineering Simulation)程式運算流程,並建立一個新的平行運算流程,以適用於GPGPU的超多核心特性。本研究將新的平行運算流程實作於美國加州大學柏克萊分校的OpenSees地震工程分析系統。測試結果顯示,本研究所測試的平行運算流程,配合GPGPU技術,得以縮短大型非線性動力分析約20%的時間。
With the advances of technology, graphics processor (Graphics Processing Unit, called GPU) with “many”-core technique subsequently became very powerful computing power. In recent years, the GPU technology was further developed for general purpose computing. This technology is known as GPGPU (General-Purpose computing on Graphics Processing Unit). The GPGPU technology brings an opportunity for large-scale nonlinear dynamic structural analyses with refined numerical models. These analyses are not commonly carried out, even if their results are more reliable and accurate. One of the reasons is their long computing time. At present, engineers and researchers commonly adopt simple and coarse numerical models to obtain analysis results quickly. Based GPGPU’s “many”-core feature, this work re-examined the procedure of finite element structural analysis, and proposed a new procedure for GPGPU computing. The new GPGPU procedure was implemented in an open-source structural analysis system named OpenSees (Open System for Earthquake Engineering Simulation). Numerical experiments were conducted to verify the performance of GPGPU OpenSees. The experimental data showed that the GPGPU enhanced OpenSees requires up to 80% of time cost of the original version in our test. The proposed procedure has potential to reduce the execution time of large-scale nonlinear dynamic structural analyses.