A special-purpose parallel computer for solving partial differential equations

The paper describes a special-purpose single instruction multiple data (SIMD) computer that implements multigrid algorithms for solving partial differential equations. The authors show that the multigrid machine is a very good alternative to commercial microprocessor-based parallel computers for solving partial differential equations, which is one of the most demanding applications. The computer obtains a very good performance/cost ratio by way of the exploitation of the innate properties in multigrid algorithms and the advantages of the massive parallelism. Moreover, it does not suffer from the problem of most special-purpose computers, i.e. the need to use a specific programming language. The computer can be programmed, from the higher abstraction level, by using an environment that automatically generates multigrid code from a problem specification. They consider this work as a first step in the building a prototype of the architecture with field programmable gate arrays (FPGAs)