ecms_neu_mini.png

Digital Library

of the European Council for Modelling and Simulation

 

Title:

Hybrid CPU/GPU Platform For High Performance Computing

Authors:

Michal Marks, Ewa Niewiadomska-Szynkiewicz

Published in:

 

(2014).ECMS 2014 Proceedings edited by: Flaminio Squazzoni, Fabio Baronio, Claudia Archetti, Marco Castellani  European Council for Modeling and Simulation. doi:10.7148/2014

 

ISBN: 978-0-9564944-8-1

 

28th European Conference on Modelling and Simulation,

Brescia, Italy, May 27th – 30th, 2014

Citation format:

Michal Marks, Ewa Niewiadomska-Szynkiewicz (2014). Hybrid CPU/GPU Platform For High Performance Computing, ECMS 2014 Proceedings edited by: Flaminio Squazzoni, Fabio Baronio, Claudia Archetti, Marco Castellani  European Council for Modeling and Simulation. doi:10.7148/2014-0508

DOI:

http://dx.doi.org/10.7148/2014-0508

Abstract:

High performance computing is required in a number of data-intensive domains. CPU and GPU clusters are one of the most progressive branches in a field of parallel computing and data processing nowadays. Cloud computing has recently emerged as one of the buzzwords in the ICT industry. It offers suitable abstractions to manage the complexity of large data processing and analysis in various domains. This paper addresses issues associated with distributed computational system and the application of mixed GPU&CPU technology to data intensive computation. We describe a hybrid cluster formed by devices from different vendors (Intel, AMD, NVIDIA). Two variants of software environment that hides the heterogeneity of our hardware platform and provides tools for solving complex scientific and engineering problems are presented and discussed. The first solution (HGCC) is a software platform for data processing in heterogenous CPU/GPU clusters. The second solution (HGCVC) is an extension version of the previous one. The cloud technology is incorporated to the HGCC framework. The results of numerical experiments performed for parallel implementations of password recovery algorithms are presented to illustrate the performance of our systems.

Full text: