A Desktop Computing System Concept for Solutions to Large Systems of Pdes

Partial differential equations (PDEs) arise in analysis of airborne structures, and the projection of natural phenomena such as severe weather and earthquakes. Unfortunately, for many useful problems, there are classes of PDEs where the computational requirements are prohibitively large and enormously time consuming. Because of this, only very small models can be resolved at an engineer's workstation and even then, this method is only acceptable when significant delays can be allowed. For larger problems, or if delays are unacceptable, the PDE solution must be obtained by submitting the work to a queue for a remote cluster of processors. However, for the largest and most complex problems, even the shortest run time on a cluster results in processing times of many days or weeks. This paper describes a High Performance Computing (HPC) desktop design that could be used to solve the highest time demanding types of PDE problems by fusing together; 1) an appropriate group of multiple reconfigurable networked computing nodes joined with 2) adapted parallel processing algorithms, where both the hardware and the algorithms are synthesized against 3) a particular PDE problem class of interest.