AN APPLICATION-SPECIFIC COMPUTER FOR LAPLACE'S EQUATION

WE PROPOSE TO DESIGN AND BUILD AN APPLICATION-SPECIFIC COMPUTER (ASPEC) TO SOLVE AN IMPORTANT CLASS OF COMPUTER AIDED DESIGN (CAD) ENGINEERING SIMULATION PROBLEMS AT FIVE TIMES THE SPEED AND ONE HUNDREDTH THE COST OF CURRENT SUPERCOMPUTERS. OUT INTENT IS TO PROVIDE O(10(9)) FLOATING POINT OPERATIONS PER SECOND (GIGAFLOPS) FOR THE SOLUTION OF O(10(6)) NODE 3-D EQUILIBRIUM-TYPE (ELLIPTIC) PROBLEMS THAT ARE UBIQUITOUS IN THE COMMERCIAL SECTOR. THE ASPEC WILL CONSIST OF A CUSTOM PIPELINED PROCESSOR BOARD AND DEDICATED, HIGH BANDWIDTH CUSTOM MEMORY, AND WILL RESIDE ON 4 TO 8 BOARDS IN A SERVER ACCESSED BY WORKSTATIONS. THE PRINCIPAL CONSTRAINT ON THE ASPEC IS THATHIGH PERFORMANCE MUST BE ACHIEVED WITH INEXPENSIVE COMPONENTS AND PACKAGING. THEREIN LIES THE INNOVATIVE PROBLEM TO BE ADDRESSED. MOTIVATION FOR THIS PROPOSAL IS THE FACT THAT A LARGE NUMBER OF COMMERCIAL OPPORTUNITIES IN CAD SIMULATION LIE FALLOW BECAUSE THE REQUIRED COMPUTER RESOURCES DISCOURAGE ENGINEERS AND THEIR MANAGEMENT IN PRIVATE INDUSTRY. THE SIGNIFICANCE OF ASPEC MACHINES IS THAT THEY OFFER A NEW DIMENSION IN THE TECHNICAL COMPUTING MARKET. GIVEN A COMPUTATIONALLY INTENSIVE APPLICATION WITH COMMERCIAL POTENTIAL, AND AN EFFICIENT, MATURE ALGORITHM FOR ITS SOLUTION, TODAY'S TECHNOLOGY ENCOURAGES "HARDWARE CODING" AS A MEANS TO MAXIMIZE COST PERFORMANCE.