Descripción
|
|
---|---|
In most of geodesic and geophysical computations matrix operations are indispensable. Especially in inverse and forward modelling problems, most of the computational cost is employed in operations as matrix multiplication, factorizations (LU, Cholesky, QR) and resolution of systems of equations. Due to the nature of the employed algorithms, as the number of data and problem dimensions increase also increases the computational cost in a greater proportion, in most cases up to a factor of N 3. Therefore, it is essential to have optimized algorithms in order to exploit all avaiable resources and minimize the required computational time. In this work the efficiency of some high performance numerical algebra libraries is analized in order to evaluate the state of the art in this field. Given a set of frequently used operations in Geodesy and Geophysics, the efficiency and the scalability of the corresponding algorithms are checked, for which matrices of different sizes are used. The tests are performed using CPU and GPU, being the latest widely used nowadays in high performance computing. Finally, a set of graphics showing the behaviour and efficiency of the checked implementations are obtained. | |
Internacional
|
Si |
Nombre congreso
|
8ª Asamblea Hispano-Portuguesa de Geodesia y Geofísica |
Tipo de participación
|
960 |
Lugar del congreso
|
Évora, Portugal |
Revisores
|
Si |
ISBN o ISSN
|
978-989-98836-0-4 |
DOI
|
|
Fecha inicio congreso
|
29/01/2014 |
Fecha fin congreso
|
31/01/2014 |
Desde la página
|
303 |
Hasta la página
|
312 |
Título de las actas
|
Proceedings da 8ª Assembleia Luso Espanhola de Geodesia e Geofísica |