Solicitamos su permiso para obtener datos estadísticos de su navegación en esta web. Si continúa navegando consideramos que acepta el uso de cookies. OK | Política de cookies | Política de Privacidad
Language: English
RSS

Follow us ...

  • Twitter FacebbokFlickrYouTube CESGA

SME Services

  • Servicios para Empresas

Annual report 2018

  • Anuario CESGA 2018

  • CESGA ICTS

The new calculation computers are based on the utilisation of multi-core processors in each one of the nodes, joined by low latency communication networks and wide bandwidth such as Infiniband. Moreover, in the last years, technologies based on graphic processors are being incorporated. In order to adapt to these changes, CESGA is conducting research on the ability to take advantage of the topology of the computers as well as in parallelization in GPUs.

The hybrid character of the architecture of Finisterrae stands out;it is a combination of  mutin-core characteristics with a NUMA shared memory architecture along with an Infiniband network interconnection, as own in the illustration below.

 

The technical report, "FinisTerrae: Memory Hierarchy and Mapping," details the influence of this architectural wealth of the matrix-vector sparse (SpMV) kernel project that is very present in scientific and engineering applications.

Following this line of research, the OPTIMISES Project has studied the effect of the techniques of resorting and the storage format of the above-mentioned kernel, executing it on GPUs and finding accelerations of up to 2.6x in the case of the resorted matrices.