Designing a graphics processing unit accelerated petaflop capable lattice Boltzmann solver: Read aligned data layouts and asynchronous communication. (May 2017)

• Record Type: Journal Article
• Authors: Robertsén, Fredrik; Westerholm, Jan; Mattila, Keijo
• Other Names: Aldinucci Marco guest-editor.; Brorsson Mats guest-editor.; D'Agostino Daniele guest-editor.; Daneshtalab Masoud guest-editor.; Kilpatrick Peter guest-editor.; Leppänen Ville guest-editor.
• Is Part Of: International journal of high performance computing applications. Volume 31:Number 3(2017)
• Page Start: 246
• Page End: 255
• Publication Date: 2017-05
• Subjects: Lattice Boltzmann -- graphics processing unit -- Titan -- asynchronous communication -- memory alignment -- data layout -- large-scale I/O -- load balance; High performance computing -- Periodicals; Supercomputers -- Periodicals; 004.1105
• DOI: 10.1177/1094342016658109
• Languages: English
• ISSNs: 1094-3420

Abstract

The lattice Boltzmann method is a well-established numerical approach for complex fluid flow simulations. Recently, general-purpose graphics processing units (GPUs) have become available as high-performance computing resources at large scale. We report on designing and implementing a lattice Boltzmann solver for multi-GPU systems that achieves 1.79 PFLOPS performance on 16, 384 GPUs. To achieve this performance, we introduce a GPU compatible version of the so-called bundle data layout and eliminate the halo sites in order to improve data access alignment. Furthermore, we make use of the possibility to overlap data transfer between the host central processing unit and the device GPU with computing on the GPU. As a benchmark case, we simulate flow in porous media and measure both strong and weak scaling performance with the emphasis being on large-scale simulations using realistic input data.