報告人：A.A. Lubrecht 教授
報告題目：High Performance Computing
報告時間：2024年9月30日下午2：30（星期一）
報告地點：泉山校區12# 6樓報告廳
主辦單位：機電工程學院、科學技術研究院
報告人簡介：
A.A. Lubrecht 教授于1987年6月畢業于荷蘭屯特(Twente)大學，獲博士學位，1987-1988,在以色列魏茨曼科學研究所讀博士后，1993年起以全職教授身份在歐洲著名摩擦學研究機構-法國里昂國立應用科學學院機械接觸與結構實驗室（LaMCoS INSA/CNRS UMR 5259）工作。他是將多重網格方法（一種高性能數值算法）引入流體潤滑領域的第一人，曾或現任多家國際摩擦學雜志副主編（如wear、Tribology International、 ASME Journal of Tribology），在國際摩擦學界具有重要影響。發表國際雜志論文250多篇，引用次數超6000次；出版的書籍《Multilevel Method in Lubrication》是摩擦潤滑領域內的一部著作。
報告摘要：
Advanced problems in Mechanical Engineering and Physics require numerical solutions, as analytical solutions can no longer be found.
With these problems becoming more and more complex, the number of unknowns or the number of degrees of freedom is quickly
approaching 10^9 to 10^10.
The current talk will describe two ways of limiting the computing time for these large problems on medium cost computers
(say 10 k$).
The first way deals with Advance Numerical Methods, in this case MultiGrid. MultiGrid allows the user to solve large and complex problems very rapidly and efficiently.
The second way deals with dedicated hardware, since two decades Graphical Processing Units (GPU) have shown amazing
price/performance characteristics. Engineers can to use the GPU power to solve huge problems in three different ways:
- Libraries exploiting GPU exist for many codes (ABAQUS, ANSYS, AUTOCAD, CATIA, FLUENT, MATLAB, .....)
- Open Acces Code
- Programming Languages (CUDA)
With the rise of Artificial Intelligence (AI), running on (NVIDIA) GPU's the development of new and more powerfull GPU's
has seen an acceleration, the spread of AI provides a big market for these GPU's, keeping prices down.
Engineers can benefit from these development to solve very large problems, by using one or BOTH tools (software & hardware).