Floating Point Systems

Floating Point Systems Inc. (FPS) was a Beaverton, Oregon vendor of attached array processors and minisupercomputers. The company was founded in 1970 by former Tektronix engineer Norm Winningstad,^[1] with partners Tom Prince, Frank Bouton and Robert Carter. Carter was a salesman for Data General Corp. who persuaded Bouton and Prince to leave Tektronix to start the new company. Winningstad was the fourth partner.^[2]

History

The original goal of the company was to supply economical, but high-performance floating point coprocessors for minicomputers. In 1976, the AP-120B array processor was produced. This was soon followed by a unit for larger systems and IBM mainframes FPS AP-190. In 1981, the follow-on FPS-164 was produced, followed by its big brother, the 264 having the same architecture. This was 5 times faster using ECL instead of TTL chips.

These processors were widely used as attached processors for scientific applications in reflection seismology, physical chemistry, NSA cryptology and other disciplines requiring large numbers of computations. Attached array processors were usually used in facilities where larger supercomputers were either not needed or not affordable. Hundreds if not thousands of FPS boxes were delivered and highly regarded. FPS's primary competition up to this time was IBM (3838) and CSPI.

Cornell University led by physicist Kenneth G. Wilson made a supercomputer proposal to NSF with IBM to produce a processor array of FPS boxes attached to an IBM mainframe with the name lCAP.

Parallel processing

In 1986, the T-Series hypercube computers using INMOS transputers and Weitek floating-point processors was introduced. The T stood for "Tesseract". Unfortunately, parallel processing was still in its infancy and the software tools and libraries for the T-Series did not ease customers parallel programming. I/O was also difficult, so the T-Series was discontinued, a mistake costing tens of millions of dollars that was nearly fatal to FPS. Possibly a few dozen T-series were delivered.

Acquisition by Cray

In 1988, FPS acquired the assets of Celerity Computing of San Diego, California, renaming itself as FPS Computing. Celerity's product lines were further developed by FPS, the Celerity 6000 minisupercomputer being developed into the FPS Model 500 series. They later became the S-MP and APP product lines of Cray Research when FPS was acquired by Cray in 1991 for $3.25 million.^[3]

The S-MP was a SPARC-based multiprocessor server (based on the Model 500); the APP an i860-based matrix co-processor array. After CRI purchased FPS, it changed the group's direction by making them Cray Research Superservers, Inc., later becoming the Cray Business Systems Division; however the S-MP architecture was not developed further, instead it was replaced by the Cray Superserver 6400, (CS6400) which was derived indirectly from a collaboration between Sun Microsystems and Xerox PARC.

After Silicon Graphics acquired Cray Research in 1996, this business unit along with the CS6400 product line were sold to Sun Microsystems. This was a great strategic mistake by SGI, as Cray were developing the Starfire system at the time, this being launched by Sun as the very successful Ultra Enterprise 10000 multiprocessor servers. These systems allowed Sun to become a first tier vendor in the large server market which Silicon Graphics never achieved.

In Usenet news group history, Steve Stevenson, a T-series owner at Clemson University started the moderated comp.hypercube news group. As hypercube architectures started to fade, he had the group renamed to comp.parallel which lasts to this day.

References

See also

• Glen Culler
• Cydrome
• Multiflow

External links

• 1986 news about FPS - Daily Journal of Commerce
• The History of the Development of Parallel Computing
• Howard Thrailkill FPS Computing: A History of Firsts
• Gordon Bell. "A Brief History of Supercomputing"