Speed addicts

The history of supercomputers - the fastest computers in the world at any particular time - has been a roller-coaster ride. This is the world of the Ferrari rather than the family saloon. Designers of flair are required for success; and failure is all too easy even with brilliant engineers at the helm.

Customers for supercomputers have often basked in the prestige of owning the fastest computer in the world. In many cases, money was not the issue if the product was available and reliable enough. Applications such as modelling nuclear explosions and weather forecasting could always use more processing power to increase the accuracy of their predictions.

By far the most well-known supercomputer designer - almost synonymous with the area as a succession of eponymous companies testify - is Seymour Cray. His relentless efforts over several decades to design the fastest computer in the world have inspired and driven generations of engineers. Cray's ability to combine theory and practice has been critical to his success; he was as happy with Boolean equations as with soldering computer circuit boards. Despite being a leader in his field, his innovation normally depended on tried-and-tested technology (typically a decade old). He preferred to allow others to make the first mistake where possible. Where he (and others) did attempt to apply unproved technology, failure often resulted. The book starts with code-breaking in the second world war and the subsequent formation of the company Engineering Research Associates in 1946, where Cray was first employed.

Only in chapter three does Cray make his appearance. He immediately made an impact as an engineer with great expertise and confidence that belied his years. Cray worked for a number of start-up companies subsequently, and in 1957 with William Norris and others he formed Control Data Corporation. With Cray at the head of the design team, the company produced what is widely considered the world's first supercomputer, the CDC6600, in the early 1960s. This took advantage of the newer and faster silicon transistor technology instead of germanium transistors. About 100 were sold at $8 million each, producing a huge profit for those days. Share prices rose from $1 in 1957 to $300 in 1964.

Like many engineers, Cray hated management interference; however, his esteemed position allowed him partially to escape the commercial pressures of management direction by moving his design team to a laboratory 80 miles from the headquarters in Minneapolis/St Paul, Minnesota. In the 1960s, even this distance required an operator-assisted long-distance telephone call, which cut unnecessary communication considerably.

Cray's normal solution to failure when it did occur was to give up completely on the design, found a new company, relocate, and start again from scratch. Often this worked, enough of the key engineers in his design team being willing to move with him.

After CDC, he founded Cray Research, which produced the renowned CRAY-1 supercomputer, using early small-scale integrated circuit technology, housed in a distinctive and iconic circular case with seating for tired engineers. Like all good designs, this was dictated by the engineering need of reducing the length of wires between modules to increase speed. Subsequent generations were less successful. Steve Chen did manage to develop a multiprocessor version, the CRAY X-MP, based on the CRAY-1 design, but this was not Cray's style. He preferred to begin with a clean sheet of paper for a new design, using the most appropriate technology available. At Cray Computer Corporation in Colorado, Cray worked on the CRAY-3, which was never successfully marketed because of inordinate delays until funding finally ran out. A perennial problem was the dissipation of heat, especially in designs using three-dimensional configurations of components where the problem was often insurmountable. The CRAY-3 included modules containing about 1,000 chips in the space of four cubic inches. In early 1996, Cray founded his last company, SRC Computers (short for Seymour Roger Cray). A parallel 512 microprocessor computer with 1 trillion floating point operations per second - 12,000 times the speed of a CRAY-1 - was planned. Unfortunately Cray suffered a fatal car accident on September 22, 1996, closing an era of supercomputer design.

The future of supercomputing has to be increasing parallelisation due to the physical limitations of sequential machines, and Cray recognised this. But that will be left to younger generations of engineers. The book tracks Cray's succession of triumphs and disappointments, and other related supercomputer developments to a lesser degree, maintaining, if not increasing, the reader's interest throughout. It makes inspirational reading for any computer engineer, and it gives an insight into the excitement possible in computing for those less well acquainted or enamoured with the subject. The technical aspects are not overwhelming; as the title suggests, this is in essence a book about people and their interactions.

Cray was a great inspirer and instigator of designs, but he relied on his colleagues to follow through his ideas to completion. He could be unforgiving of those who made mistakes or slow progress, but he was very protective of colleagues whom he valued. At one point he even worked for the minimum wage of $1.25 an hour to avoid any of his design team being laid off. When working on a problem, he required long periods of silence for concentration, a demand that his co-workers learned to respect. He could lay down and normally obtain his requirements because of his huge ability and reputation. As Fortune magazine said of Cray in April 1966, "There is no doubt that, in a field where genius is almost taken for granted, he is a towering figure." This book is a fitting tribute to the genius of Cray and all of his colleagues over the years.

Jonathan Bowen is lecturer in computer science, University of Reading.