The New Whiz Kid : Radical Computer Design Speeds Up Calculations and Competition
Ask Rodney Hornstein to explain what he means by a “massively parallel computer,” and he is ready with a very non-technical reply.
“It’s a bit like a chorus line,” explained Hornstein, president of Active Memory Technology, an Irvine computer company. “The different parts all do the same thing at the same time.”
Active Memory Technology is one of a handful of small U.S. companies that are developing super-fast computers with a radically new design. This new generation of machines promises to speed up many complex computing tasks, such as simulating the airflow around an aircraft wing. At the same time, these machines will cost far less than the largest computers now in use.
Known in industry jargon as “parallel processors,” these computers link anywhere from dozens to thousands of different processors within the computer to simultaneously tackle a computing problem. The purpose of having many parts of the computer work on the same problem at once is simple: a faster solution.
Traditional computer designs use a “serial” approach in which a single processor works out the problem one piece at a time. Serial, or sequential, technology has been the basis for most computer designs for the past 30 years.
AMT calls its computer “massively parallel” because of the unusually large number of separate processors it uses. The company’s first computer uses 1,024 processors packed into a box about the size of a two-drawer file cabinet. And AMT introduced a new model last September that uses 4,096 processors.
AMT says its parallel machines, which cost $120,000 and up, can exceed the performance of supercomputers, today’s most powerful machines. Supercomputers, which utilize a small number of very powerful processors, carry price tags in the millions.
Hornstein concedes that parallel architecture won’t be better than supercomputers in all situations. While parallel processors are well-suited for certain types of computing tasks, such as processing radar signals, modeling complex scientific experiments and detecting potential flaws in sophisticated computer microchips, supercomputers are particularly good in applications such as weather forecasting and automobile design.
Researchers at Washington University in St. Louis are using an AMT computer in a federally funded project to develop a faster method for processing images from sophisticated medical devices known as PETT, for positron emission transaxial tomography scanners. The devices are primarily used for brain scans.
Robert E. Morley, an associate professor in the university’s engineering department, said the purpose of the project is to sharply reduce the processing time and improve the quality of PETT images.
“The computations for producing these pictures are very complex and require lots of computing time on a conventional computer,” Morley said. “With a parallel processor, we can get the pictures back in a couple of minutes rather than days.”
Privately held AMT was formed in 1986 when a group of venture capitalists bought the business from ICL Ltd., a large British computer maker. ICL spent $50 million over 12 years developing the technology used in AMT’s machines before deciding that the operation didn’t fit into its business plans.
ICL still owns 17% of AMT’s stock. About 13% of the stock is owned by AMT employees, including co-founders Hornstein and Geoffrey Manning, a British nuclear physicist, and various venture capital firms own the rest.
AMT sold $6.5 million worth of its machines in 1988 and is hoping to triple sales in 1989. The firm was unprofitable last year, but managed to post “break-even” results in the final quarter, Hornstein said.
The company employs 45 people at its Irvine headquarters, which also is the site of its engineering and manufacturing operations. It also has a 45-person software development staff in England.
AMT’s Hornstein said the company’s goals are to reach $50 million in sales by 1991, make a “reasonable” profit and sell its stock to the public.
Achieving those goals won’t be easy.
The parallel-processing field has become very competitive, and the market has not grown as quickly as some forecasters predicted several years ago.
Some analysts projected the market would hit $500 million in 1988 and $1 billion by 1990. But only $330 million worth of parallel machines were sold in 1988, according to Dataquest, a San Jose market research firm. Dataquest projects that sales could reach $1 billion by 1992.
Christopher Willard, a Dataquest analyst, said that although sales of parallel computers have been growing steadily, revenue has not grown as quickly because the cost of the machines has been dropping.
“That can be interpreted as a good sign to the extent that people are buying these systems, which will provide a future customer base,” Willard said. “The downside is that there is a lot of price competition as a result of too many companies going after the same market. A lot of these companies are hurting, and there may be a fair shakeout in the market.”
AMT sold 36 computers in 1988, mostly in the United States and Europe. About half of those sales were to universities, with defense contractors and research institutions accounting for most of the rest.
One of the challenges for parallel computer makers is overcoming customers’ reluctance to adopt what for many of them is an unfamiliar technology.
“The market studies say we must show customers that we can improve the price and performance by a factor of 10 times before they are willing to consider a new technology,” said Bill Hogan, vice president of marketing at Maspar Computer, a Silicon Valley company that plans to introduce a massively parallel computer later this year.
Hogan gives AMT high marks for its products and its efforts to carve out a place in the emerging market. “We want them to be very successful,” he said. “All of us in this business believe the future of computing lies in parallel processing.”
Maybe so, but Hornstein concedes that AMT needs to land a major order soon to get its business into the black and keep its investors happy. The defense business offers the best prospect for a big order.
“Our biggest problem is the length of time that these defense program deployments take,” Hornstein said. “We’re going to have to depend on sales to academic institutions in the early years in order to have the income to wait for the bigger defense orders.”
AMT has several orders from defense contractors, but company officials won’t say much about them, explaining that the programs are classified.
Hornstein said the company hopes to hear by late 1990 whether one of its defense customers will use AMT equipment in a large weapons system. The sale would provide a major financial boost for the company, he said.
“We’re either going to be very successful, or we’re going to just die,” Hornstein said. “We don’t have the luxury of being in the middle.”
