An Application-Specific Computer for Laplace's Equation

建议研究一种特定于应用的计算机体系结构，以解决一类基于拉普拉斯方程的计算机辅助设计模拟计算。其目标是以当前超级计算机的五倍速度和百分之一的成本来解决班级中的典型问题。其目的是为解决0(106)个节点的三维椭圆问题提供每秒0(109)次浮点运算。正在进行的研究是：(1)分析和绘制迭代拉普拉斯求解器的算术运算和内存访问；(2)设计相应的多FPU流水线处理器；(3)确保现有存储板设计的适用性，并确定维持流水线所需的矢量存储端口(板)的数量；(4)编写流水线模拟器；以及(5)规划流水线处理器的线绕制造

The proposal is to investigate an application-specific computer architecture to solve a class of computer aided design simulation computations based on the Laplace equation. The goal is to solve typical problems in the class at five times the speed and one hundredth the cost of current supercomputers. The intent is to provide 0 (109) floating point operations per second for the solution of 0(106) node 3-D elliptic problems. Research being conducted is: (1) analyze and diagram the arithmetic operations and memory accesses of the iterative Laplace solver; (2) design the corresponding multi-FPU pipelined processor; (3) insure suitability of an existing memory board design and determine the number of vector memory ports (boards) necessary to sustain the pipeline; (4) program a simulator of the pipeline; and (5) plan wirewrap fabrication of the pipelined processor