Seymour Cray's gone, but not forgotten

OCTOBER THE fifth marked the anniversary of the untimely death of Seymour Cray, a man who is generally regarded as the father of supercomputing.

This seems an appropriate time to look back at his achievements and consider his impact on computing as we know it today.

Cray was born in Wisconsin in 1925 and served in the US army during World War II. On discharge at the end of the war he received a Bachelor of Science in Electrical Engineering and a year later a Masters degree in Applied Mathematics. Following the advice of a professor, Cray joined Engineering Research Associates (ERA), a one year old company that was one of the first to use digital circuits in their work for the US Navy.

His early work was on the 1100 series computers from ERA. Cray had contact with Von Neumann but with computing was in its infancy, many problems had not been explored and he had to rely on his own research and invention.

After working as a team member on two computers at ERA Cray was given the task of designing the next computer, the 1103. Using the earlier models as a base, Cray designed for simplicity and speed, and he used RISC principles even before the name had been invented.

In the middle of the 1950s, ERA was purchased by Remington Rand, followed later by Sperry, then Burroughs and most recently by Unisys. Remington changed the emphasis from scientific computers to computers for commercial businesses. Eventually Cray became dissatisfied with this direction and in 1957 he resigned to join Bill Norris and start Control Data Corporation (CDC) where they returned to their roots by concentrating on producing systems for the scientific and mathematical world.

By 1960, aged only 34, Cray had established a reputation for designing high performance computers. He had finished the design of the Control Data 1604, the first fully transistorised computer, and had begun work on the first major supercomputer the CDC 6600.

Over time, the firm became aware of the limitations of single processor systems and in 1970 released the first four-processor machine, a forerunner of many current systems.

Cray tended to work alone on his designs. He would draw on the complaints of customers and develop solutions to their problems. It was only when the basics of a design were defined would Cray call on his support staff to tie it all together and build the computer.

One of CDC's first operating systems was called SCOPE and this gives a strong hint that Cray's work was extending beyond hardware.

By 1972 the writing was on the wall for CDC because its market was diminishing. Other companies were now on the scene, many with quite reasonable computers for their target markets. Rather than stay and risk seeing CDC fade like ERA had done earlier, Cray left CDC and started his own company, Cray Research Inc., to continue work with high performance machines.

He used the new technology of integrated circuits and vector register methods to build the Cray 1, first released in 1976. Just six years later Cray had multiprocessor machines which ran at 1GHz, a far cry from Digital's Vax series ran at 10Mhz - speedy for the time. This was also years before the appearance of the first personal computers.

Following the Cray-1 came the XMP, a machine that I worked on for several months.

The XMP was an extraordinary machine in many ways. It used vectorisation, integrated rather than discrete components and semiconductor memory rather than magnetic core.

The processor consisted of a series of discrete boards, roughly 15cm square, each with a small number of components. The engineers actually used torque screwdrivers to screw the holding bolts only to a certain pressure. The whole machine had the appearance of a split and displaced cylinder because this reduced the length of the connecting wires. Around the cylinder was a padded area which was sometimes used as a seat but actually hid the piping of the Freon-based cooling system.

The Cray processor was one of the first to work with multiple instructions. Only one instruction could be started at any clock cycle but once that instruction was started the next instruction could start on the next cycle. This isn't easy -- different instructions require a different number of cycles to complete.

Compilers for high performance systems were a very closely guarded secret because they were an integral part of the package but despite this they were very user-friendly. The COS Fortran compiler produced an optional listing which described how processing loops had been unrolled, or vectorised, how constants had been moved and it made suggestions as to how the code could be modified to optimise it even further.

Cray was probably also involved in the creation of the COS operating system which was similar to SCOPE on the CDC systems. In later machines its version of Unix, Cray UNICOS, was more common.

The Cray 2 arrived in 1985 - a very successful machine which continued the performance advantage that the Cray 1 had shown over contemporary machines. This new machine had up to 4GB memory and up to eight processors.

In 1989 Cray started a new company called Cray Computer Corporation to develop the use of gallium arsenide as a replacement for silicon. He devoted most of his time to this new company but retained an interest in Cray Research as it pursued a slightly more conventional line.

After the release of Cray 2 came the YMP in 1988, the C90 and YMP-EL in 1991 and then in 1993 the T3D, a massively parallel machine based on Digital's Alpha processor.

The struggle to harness the advantages of gallium arsenide proved difficult and it was not until 1993 that the only Cray-3 ever built was handed over to its potential customer for trials and further tests by its developers. It was switched off in March 1995 just one day after CCC filed for bankruptcy.

In 1995 Seymour Cray started a new business named, SRC Computers, with an aim to continue his work in supercomputing. At the time he said: "We think we'll build computers, but who knows what kind or how. We'll talk it over and see if we can come up with a plan."

On 22 September 1996 his Jeep Cherokee was involved in a collision which left him with major head injuries and he died on the 5th October following.

In an address to the Supercomputing conference held in November of that same year, Charles W. Breckenridge of SRC stated, "No one in history in Seymour's field has accomplished the consistent successes that he recorded during his lifetime. He dedicated his entire career to the design and development of large-scale, high performance systems for science and engineering. He often said that he felt he was put on earth to do that job."

Cray was often described as a reclusive man but this was hardly a fair description. He was deeply interested in his work and the work of others. He was also a great listener and explorer of ideas. He was also active in many sports but as an individual rather than part of a team. He basically shunned the limelight and his only vanity appears to have been in the names that he chose for his companies.

Finally, Cray's commercial legacy lives on. After his death, Cray Research was sold to Silicon Graphics, then in 1999 to Teradata which renamed the company to Cray, Inc. µ

More
There are hundreds of web references to Seymour Cray. Some that I found particularly interesting are the following.

A 10-page PDF document from 1978 describing the Cray-1
Tribute from Charles W. Breckenridge
Article by Jason Pepper
Homepage of Cray-cyber.org
The historic Cray systems page of the website of Cray.com
An obituary from a UK writer
Start of Gordon Bell's lecture about Seymour Cray
A long transcription of an 1995 interview with Seymour Cray