首页> 外文会议>Recent researches in communications and computers >Simulation of Multiphase Porous Media Flows on High-Performance Hybrid Computing Systems
【24h】

Simulation of Multiphase Porous Media Flows on High-Performance Hybrid Computing Systems

机译:高性能混合计算系统上多相多孔介质流的模拟

获取原文
获取原文并翻译 | 示例

摘要

Nowadays, the rapid growth in the computer performance is mainly achieved due to the use of hybrid architectures including the multi-core CPUs and graphics accelerators. However, such architectures cause serious difficulties in their efficient employment and make new demands to the software development. One of the most promising approaches to construct the efficient applications for such systems consists in the creation of computational algorithms with a simple logical structure. This approach has been used at the development of the cross-platform unified parallel program library for modeling such large-scale processes in the underground space as given by the contaminant infiltration into the soil, oil recovery problems and so on. The library is written in C++ using the MPI and CUDA technologies. It consists of calculation, communication as well as control modules and allows computations of 2D and 3D problems with double precision, which can be realized on both the GPU and the NUMA cluster multi-core CPUs, the access to different memory types is optimized. To verify the developed parallel programs, some test calculations in the infiltration and oil recovery problems have been performed on the hybrid supercomputer with 100 TFLOPS peak performance.
机译:如今,计算机性能的快速增长主要是由于使用了包括多核CPU和图形加速器的混合体系结构。但是,这样的体系结构在有效使用上造成了严重的困难,并对软件开发提出了新的要求。为此类系统构建有效应用程序的最有前途的方法之一是创建具有简单逻辑结构的计算算法。这种方法已用于跨平台统一并行程序库的开发,该模型用于对地下空间中的大规模过程进行建模,这些过程是由污染物渗入土壤,采油问题等引起的。该库使用MPI和CUDA技术以C ++编写。它由计算,通信和控制模块组成,并允许以双精度计算2D和3D问题,该问题可以在GPU和NUMA集群多核CPU上实现,并优化了对不同内存类型的访问。为了验证已开发的并行程序,已经在峰值性能为100 TFLOPS的混合超级计算机上对渗透和油回收问题进行了一些测试计算。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号