Developing a High Performance Software Library with MPI and CUDA for Matrix Computations
Journal Title: Computational Methods in Social Sciences - Year 2013, Vol 1, Issue 2
Abstract
Nowadays, the paradigm of parallel computing is changing. CUDA is now a popular programming model for general purpose computations on GPUs and a great number of applications were ported to CUDA obtaining speedups of orders of magnitude comparing to optimized CPU implementations. Hybrid approaches that combine the message passing model with the shared memory model for parallel computing area solution for very large applications. We considered a heterogeneouscluster that combines the CPU and GPU computations using MPI and CUDA for developing a high performance linear algebra library. Our library deals with large linear systems solvers because they are a common problem in the fields of science and engineering. Direct methods for computing the solution of such systems can be very expensive due to high memory requirements and computational cost. An efficient alternative are iterative methods which computes only an approximation of the solution. In this paper we present an implementation of a library that uses a hybrid model of computation using MPI and CUDA implementing both direct and iterative linear systems solvers. Our library implements LU and Cholesky factorization based solvers and some of the non-stationary iterative methods using the MPI/CUDA combination. We compared the performance of our MPI/CUDA implementation with classic programs written to be run on a single CPU.
Authors and Affiliations
Bogdan Oancea, Andrei Tudorel
Keywords
Related Articles
- EP ID EP88071
- DOI -
- Views 181
- Downloads 0
An overview of higher education at the European level
Higher education has a key role in the new knowledge society. In Europe there are over 4,000 higher education institutions of which 3300 are in EU27 countries and over 20 million students. EU countries' governments and h...
Simulating the binary variates for the components of a socio-economical system
Often in practice the components Wj of a sociological or an economical system W take discrete 0 - 1 values. We talk about how to generate arbitrary observations froma binary 0-1 system B when is known the multidimensiona...
The Likert scale analysis using parametric based Structural Equation Modeling (SEM)
The Likert scale is commonly used in survey research using primary and secondary data to measure the respondent attitude by asking insofar to which they agree or disagree with a particular questions....
Large-scale natural disaster analysis in European transition countries
The first aim of this study is to identify possible natural disasters and catastrophes and to summarize the state of economic literature in the domain. Further are analyzed the methodologies used for estimating the econo...
Human resources in the economic crisis
This paper is meant to be an extension of our studies published over recent years, which were meant to seek some answers regarding the existing cause - effect relationship between the economic, financi...