基于多GPU的ReaxFFMD并行策略的设计和实现
详细信息 查看官网全文
- 英文论文题名：Algorithms of multi-GPUs enabled reactive molecular dynamics(ReaxFF MD) program
- 论文作者：郑默 ; 李晓霞 ; 郭力
- 英文论文作者：Mo Zheng ; Xiaoxia Li ; Li Guo ; State Key Laboratory of Multiphase Complex Systems ; Institute of Process Engineering ; Chinese Academy of Sciences
- 年：2016
- 作者机构：中国科学院过程工程研究所多相复杂系统国家重点实验室;
- 论文关键词：反应分子动力学ReaxFF MD ; MPI-GMD-Reax ; GPU ; 大规模模拟
- 会议召开时间：2016-07-01
- 会议录名称：中国化学会第30届学术年会摘要集-第二十五分会：化学信息学与化学计量学
- 语种：中文
- 分类号：O643.12
- 学会代码：ZGHY
- 会议名称：中国化学会第30届学术年会-第二十五分会 ; 化学信息学与化学计量学
- 会议地点：中国辽宁大连
- 主办单位：中国化学会
- 学会名称：中国化学会
- 页数：1
- 文件大小：141k
- 原文格式：D
- 会议级别：全国
摘要
化学反应分子动力学ReaxFF MD~1方法可连续描述化学键的断裂和生成,且不必预先定义反应路径,成为有潜力可从分子层次认识复杂体系反应机理的新途径。针对ReaxFF MD可模拟的时空尺度受限~2,难以应用于大规模体系(>50,000原子)反应模拟的问题,本文提出了基于图形处理器GPU和MPI消息传递编程模型的ReaxFF MD并行策略,并创建了跨节点、多GPU并行加速的ReaxFF MD程序系统MPI-GMD-Reax。MPI-GMD-Reax充分利用了GPU强大的流处理器、CUDA支持大量线程级并行及MPI高效灵活的特性,进一步提升了ReaxFF MD的整体性能和运行规模。保守估计,在四节点八GPU卡的多GPU(K20C)运行中可将ReaxFF MD的模拟规模扩大至400,000原子。
ReaxFF MD provides a new and promising approach for molecular simulation of complex reactive system the underlying with chemical reactions.In order to reach spatio-temporal scales of nanometers and nanoseconds for large-scale system(>50,000 atoms),the paralleled implementation of ReaxFF MD on multi-GPUs(MPI-GMD-Reax) is developed based on the GMD-Reax program by taking advantage of message passing interface(MPI).The implemented algorithms of MPI-GMD-Reax have significantly improve increase the computational performance and system scale of ReaxFF MD.MPI-GMD-Reax has a capability for simulating a much more complex system with around 400,000 atoms when running on four processors with 8 K20 C GPUs.
ReaxFF MD provides a new and promising approach for molecular simulation of complex reactive system the underlying with chemical reactions.In order to reach spatio-temporal scales of nanometers and nanoseconds for large-scale system(>50,000 atoms),the paralleled implementation of ReaxFF MD on multi-GPUs(MPI-GMD-Reax) is developed based on the GMD-Reax program by taking advantage of message passing interface(MPI).The implemented algorithms of MPI-GMD-Reax have significantly improve increase the computational performance and system scale of ReaxFF MD.MPI-GMD-Reax has a capability for simulating a much more complex system with around 400,000 atoms when running on four processors with 8 K20 C GPUs.
引文
[1]van Duin,A.C.T.;Dasgupta,S.;Lorant,F.;Goddard,W.A.,J.Phys.Chem.A.2001,105(41):9396-9409.
[2]Zheng,M.;Li,X.;Guo,L.,J Mol Graph Model 2013,41:1-11.
[2]Zheng,M.;Li,X.;Guo,L.,J Mol Graph Model 2013,41:1-11.