Tokyo, Japan -- As reported in Nikkei English News, the Real World Computing Partnership said it has developed a massively parallel processing system that links a large group of personal computers into a network for extremely fast parallel processing. A network comprising 32 PCs offers the processing power of a low-end supercomputer, but for just 20% to 40% the cost. The organization will hurry to develop a practical technology that can be used to build inexpensive yet powerful computer systems for such applications as scientific calculations and semiconductor circuit design. Massively parallel processing systems utilize a number of microprocessors operating in concert to more efficiently crunch through complicated programs at high speed. The processors can be physically linked into a cluster, or they can be connected over a network. The new system is of the latter type, comprising 32 circuit boards, each with an Intel processor and main memory, connected together via a high-speed network. The network lines can accommodate data transfer speeds of 160 megabytes per second. The PCs act as terminals through which to access this massively parallel processor. When the user inputs instructions from a terminal, the program is divided up and distributed among the processors for parallel computation, then the results are sent back to the originating terminal. Tests at the organization's research center in Tsukuba suggest the system can process data at the extremely fast rate of approximately 3 billion floating point operations per second, or 3 gigaFLOPS.
Linked PC’s Deliver Power of Low-End Supercomputer
October 10, 1997