The Seymour Cray Era of Supercomputers (Review)
The Seymour Cray Era of Supercomputers: From Fast Machines to Fast Codes is a technical and business history of the roughly three-decades when Seymour Cray dominated the development of a class of computer called the “supercomputer”. The book covers the development of the major supercomputer models, the technical decisions and trade-offs involved, and changes to the market. The book ends with SGI’s purchase of Cray’s assets and the transition to massively parallel processing.
The Seymour Cray Era of Supercomputers: From Fast Machines to Fast Codes. By Boelie Elzen and Donald MacKenzie. ACM Books. ISBN 979-8-4007-1369-9. DOI 10.1145/3705551.
Early on, computer designs were bifurcated between the domains of “business computing” and “scientific computing.” Business computing was almost exclusively fixed-point, dealt with categorical and string data, often I/O bound, and had a broad base of less technical users with often similar problems (e.g. payroll). Scientific computing heavily used floating point, dealt with complex numerical data such as matrices and chains of differential equations, often CPU bound, and featured a very small base of highly technical users with specialized and often unique problems (e.g. national laboratories).
In 1964, this bifurcation was exemplified by the release of two products. IBM launched the IBM 360, a new computing platform that came to dominate business management. Control Data Corporation (CDC), co-founded by Seymour Cray, released the CDC 6600. While IBM sold more than 1,000 in its first month, the CDC took a decade to sell 100. However, the 6600 claimed the moniker “fastest computer in the world” and only relinquished the claim to its successor, the 6700. Although IBM could take solace in finding greater monetary success in a much larger market, Thomas J. Watson Jr. was mortified that IBM had lost the performance crown to a tiny team with far fewer resources.
Over ten chapters, the authors chronicle the technical developments that kept supercomputers at the performance edge for the next three decades. The book covers the development of the CRAY-1 and the other major models that came out of Cray’s companies, as well as CDC, IBM, Texas Instruments, and Fujitsu efforts to compete in the area. Supercomputers expanded from a small pool of government and university customers into industries such as aerospace, petroleum, and automotive and the authors discuss each industry’s “hero problems” and the computer vendors approached the new markets. The book ends as economics shift and massively parallel processing becomes the dominant approach.
Technology development is inherently risky and the book goes into detail on the many false starts and near misses. Although the focus of the book is usually on hardware and physical concerns such as wire propagation delay, the book pays attention to software concerns in parallel to market changes. Early on, supercomputer customers wrote almost all of their own software but, as the market grew, customers expected more software support from vendors. The latter half of the book includes discussion of operating system development, evolution of compilers (mostly Fortran) and their interplay with instruction sets, and specialized tools for specific industries (e.g. computational fluid dynamics). I was unaware of the early resistance to Unix and the concern that it was “slow.”
As a criticism, the text ignores aesthetics and personalities of the people involved. We learn very little about Seymour Cray or any other designer’s personality or what it was like to work for them. Yet, supercomputers were deliberately designed to beautiful or at least evocative. This sterility of the text and its approach misses an important aspect of why people were drawn to supercomputers and their impact in the culture.
The book is available digitally, in paperback, and hardbound. I read the paperback version which is printed on demand. The text and photos were crisp and highly readable, although the ink did bleed through the pages. Additionally, the left-hand internal margin was tighter than comfortable, although not so tight as to obscure any of the text.
Recommended for this interested in the history of scientific computing, computer architecture, and high-value/low-volume business strategy.