Applications of heterogeneous computing in computational and simulation science. (1st January 2013)

• Record Type: Journal Article
• Authors: Domanski, Luke; Bednarz, Tomasz; Gureyev, Timur; Murray, Lawrence; Huang, Bevan Emma; Nesterets, Yakov; Thompson, Darren; Jones, Emlyn; Cavanagh, Colin; Dadong, Dadong; Vallotton, Pascal; Sun, Changming; Khassapov, Alex; Stevenson, Andrew; Mayo, Sheridan; Mayo, Mayo; George, Andrew W.; Taylor, John A.
• Is Part Of: International journal of computational science and engineering. Volume 8:Number 3(2013)
• Page Start: 240
• Page End: 252
• Publication Date: 2013-01-01
• Subjects: heterogeneous computing -- GPU computing science -- high performance computing -- HPC -- scientific computing -- parallel programming -- graphics processing units -- GPUs -- cluster; Computer science -- Mathematics -- Periodicals; Computer simulation -- Mathematical aspects -- Periodicals; Computational intelligence -- Periodicals; 004.015105
• Languages: English
• ISSNs: 1742-7185

Abstract

This paper presents a summary and overview of heterogeneous algorithms and applications developed by the Commonwealth Scientific and Industrial Research Organisation (CSIRO) for solving practical and challenging science problems such as: 1) high-content analysis of brain cell images for medical research and drug discovery; 2) genetic analysis of complex experimental designs for crop breeding; 3) solving computational fluid dynamics problems; 4) deconvolving 3D images from microscopy and medical imaging; 5) reconstructing large 3D computed tomography (CT) images from medical and materials science; 6) quantifying uncertainty in complex environmental models. Applications discussed utilise GPUs and multi-core CPUs on a scale ranging from single desktop workstations through to large GPU clusters. Results demonstrate that both types of system can successfully accelerate a variety of practical science applications. We have seen significant gains in productivity and opportunity as a result of leveraging GPUs, tackling computational problems in which execution time was previously infeasible.