As a rule, the world's largest computer company takes great pains to downplay the speculation that swirls around its plans, ostensibly for fear of jeopardizing sales of its existing machines.
But departing from its usual practice, International Business Machines Corp. is publicly portraying itself as a future leader in what many scientists believe is emerging as the next generation of computer architecture: parallel processing.
Parallel processing, simple in concept but difficult in execution, refers to interconnecting multiple processors to work on a problem and solve it faster (see story, Page K2).
"We're very interested in almost every form of parallelism," said Ralph E. Gomory, IBM's senior vice president and director of research. "Our resources in the area are substantially higher than they were just a few years ago; we now have more than 50 people working on parallel processing in our research facilities."
IBM, which has more of a reputation as a computer marketer than a technological innovator, apparently feels that these new computer-design approaches, which could mean significantly faster computations at a fraction of the cost, offer market opportunities that the $50-billion-a-year giant can't afford to pass up.
It also puts IBM's competition on notice that the company will not allow others to exploit new technologies to gain a share of the market at IBM's expense. At the same time, IBM is letting its customers know that, if they're patient, IBM will satisfy their craving for state-of-the-art parallel-processing power.
"There's a certain unavoidable quality to it," said IBM's Gomory. "People are driven to compute at speeds faster than the speed at which one can speed up a single central processing unit. I think there are very strong forces that push you towards multiple computers, not because it's easy but because you have to. The outlook for parallelism is very bright because, at a certain level, it will become the only way to get speed."
IBM recently revealed that it has been working on the Research Parallel Processor Prototype (RP-3), an experimental parallel processing machine, in cooperation with New York University's Courant Institute. IBM also said it will offer a special "array processor" that provides limited high-speed parallel-processing capabilities for its new 3090 "Sierra" mainframe computers. These are ways that IBM can move into the burgeoning high-speed supercomputer market now dominated by Cray Research Inc.
"We're very interested in the concept of supercomputers," said Jack D. Kuehler, IBM's senior vice president in charge of manufacturing and development. "It's a result of our customers' increasing interest in getting supercomputer-like functionality."
Because of its size and spending power -- $4 billion plus, annually -- on research and development, IBM is able literally to spread its chips across the table in betting on parallel processing designs. Since it is not yet clear which will prove most fruitful, such hedging, which smaller competitors cannot afford, gives IBM an important advantage.
"What is the right single direction?" Kuehler asked. "I don't think there is a right single direction. There are multiple paths, and IBM has elected to work on every single one of them and understand the benefits and the disadvantages of each one. That's why we have been reluctant to raise our hand and say this one is the supercomputer direction of the future."
Currently, IBM is hard at work developing an "interconnection mechanism," which Gomory said functions "like a high-speed switch," that will enable these processors to communicate and exchange the required problem-solving information.