Large increases in cost with questionable increases in performance
can be tolerated only in race horses and women.” Lord Kelvin
Seymour Cray worked at Engineering Research Associates on computer design for the US Navy. Cray focused on cooling technologies and magnetic amplifiers and came to prominence for his design work on the ERA 1103, the first commercially-successful scientific computer. He stayed there when ERA was acquired by Remington, then Sperry, and eventually UNIVAC.
1957 Cray co-founded the Control Data Corporation where he was responsible for the design of the CDC1604 in 1960, an enhanced ERA 1103. He went on to produce the CDC6600, the very first supercomputer.
1968 He helped to design the CDC 7600, a second-generation refrigerated supercomputer. Until 1975 it was claimed to be the fastest computer in the world at 10-megaflops. A megaflop is a speed of 10⁶ floating point operations/ second.
The CDC 7600 ran a million times faster than a modern calculator; of course a calculator needs only to operate a little faster than its human operator and at this time they usually achieved around 10 FLOPS.
From the late 1960s to early 1970s Cray was involved in developing the CDC8600, essentially four 7600s put into one case using parallel computing. But this ran into technical difficulties. At the time the organisation was also over extended on the CDC STAR-100.
Cray moved on in 1972 to found Cray Research where he developed the Cray 1. In 1976 the first Cray 1 was installed at the Los Alamos National Laboratory on a six-month trial basis. In 1977 the National Center for Atmospheric Research was t he first formal customer, paying just under $8m for the computer and $1m for the software.
Cray 1 was forecast to have a global sales potential of twelve units, but in fact sold over eighty at between $5m and $8m each. It was a 64-bit system planned to achieve 160 FLOPS; with vector instruction software innovations it could peak at 250 FLOPS.
Charles Simonyi joined NASA’s Ames Research Center to work on the Illiac IV. This supercomputer used 256 processors in parallel with software called vector processing. Completed in 1976, and well over budget, it could not match the Cray 1.
In 1996 Silicon Graphics Inc merged with Cray Research for $740m to work on the CS6400 SPARC-based supercomputer.
By 2008 IBM’s Roadrunner was installed at the Los Alamos National Laboratory at a cost of $133m achieving 1.026 petaflops (1015 FLOPS) and designed to eventually achieve 1.7 petaflops.
The Cray XT5 Jaguar installed at the Oak Ridge National Laboratory in 2009 was able to sustain 1.759 petaflops to become the fastest supercomputer.
An alternative to faster and parallel central computers emerged as distributed computing. The Berkeley Open Infrastructure for Network Computing was developed for use with the SETI@home project, Search for Extra-Terrestrial Intelligence. It is an open-source approach with users pledging their unused PC time to collaborative scientific projects. BOINC has 1m+ users from 253 countries – SETI has yet to record any ET transmissions!
1999 BOINC achieved an average 5.4 petaflops and became the basis for volunteer computing projects. More than twenty such projects are operating today.