Abstract
High resolution reservoir modeling has been a valuable tool for reservoir simulations of the giant fields managed by Saudi Aramco; the world leader in high performance reservoir simulation studies utilizing state-of-the-art computer facilities. Three of its current computing clusters used for reservoir simulations are in the list of top 5001 computing platforms in the world. The computing center in Saudi Aramco currently hosts systems with computing power over 300 teraflops dedicated for reservoir simulation activities. This is complemented with over 750 terabytes of data storage. POWERS2 , the Saudi Aramco in-house developed parallel simulator, is used for most simulation studies on these computing platforms. It has been ported to many parallel systems, such as the Thinking Machine, IBM Nighthawk, Linux clusters, Blue Gene, etc. The simulation environment is being continuously enhanced to support studies containing complex wells, coupled surface facility, etc. The algorithm complexity and memory contentions in the underlying software and hardware present challenges in efficient use of parallel computing platforms. In this paper we review various benchmark efforts conducted in Saudi Aramco to determine performance of the latest computational hardware and software tools for computations of large simulation models. We study strength and limitations of emerging architectures, such as graphics processing units (GPUs), to solve various important computation and communication intensive building blocks in POWERS.
