论文部分内容阅读
本文叙述了一种专门为解决理论物理方面的重要问题而设计的并行处理器。其最终设备拥有256个节点。这些节点以SIMD的方式进行锁步(lock step)运行,并具有每秒40亿次22位浮点操作的运算能力。每一个节点均由一个Intel 80 286/287微处理器控制,每个节点拥有160K的存貯位和一个流水线式的、可编微程序的执行浮点乘法和加法的运算器。为了便于利用任何两个相邻节点上所貯存的数据进行向量化运算,这些节点以二维长方网格方式互连。
This article describes a parallel processor designed to address important theoretical and physical issues. Its final device has 256 nodes. These nodes operate in lockstep SIMD mode and have the computational power of 4 billion 22-bit floating-point operations per second. Each node is controlled by an Intel 80 286/287 microprocessor, each node has a 160K memory bit and a pipelined, microprogrammable floating-point multiply and add operator. In order to facilitate the vectorization operation using the data stored on any two adjacent nodes, these nodes are interconnected in a two-dimensional rectangular grid.