Development Of A Low-Cost Hybrid Massively Parallel Computing System

Principal Investigator: A/P Bertil Schmidt

Starting Date: 1999-08

Duration: 3 years

Funding Source: ISAVISION GmbH

Overview:

There are currently two forms of low cost high performance parallel computing: �Medium grain, SPMD-type (Single program, multiple data stream), computer clusters �fine grain, SIMD-type (Single instruction stream, multiple data stream), massive parallel systems The aim of this project is to set up and research a hybrid system that combines both of these low cost paradigms. This concept has the potential of outperforming supercomputers in special areas of applications � in particular in areas that lent themselves to data-parallelism, i.e. large problems can be handled by partitioning and assigning data blocks to different components of the system. The objective of this project is to provide low cost solutions to problems that are up to now within the realm of expensive supercomputers.