Meanwhile, at the new Chippewa Falls lab, Seymour Cray, Jim Thornton, and Dean Roush put together a team of 34 engineers, which continued work on the new computer design. One of the ways they hoped to improve the CDC 1604 was to use better transistors, and Cray used the new silicon transistors using the planar process, developed by
Fairchild Semiconductor. These were much faster than the
germanium transistors in the 1604, without the drawbacks of the older mesa silicon transistors. The speed of light restriction forced a more compact design with refrigeration designed by Dean Roush. In 1964, the resulting computer was released onto the market as the
CDC 6600, out-performing everything on the market by roughly ten times. When it sold over 100 units at $8 million ($ million in dollars) each; it was considered a
supercomputer. The 6600 had a 100ns,
transistor-based
CPU (Central Processing Unit) with multiple asynchronous functional units, using 10 logical, external
I/O processors to off-load many common tasks and
core memory. That way, the CPU could devote all of its time and circuitry to processing actual data, while the other controllers dealt with the mundane tasks like punching cards and running disk drives. Using late-model
compilers, the machine attained a standard mathematical operations rate of 500
kiloFLOPS, but handcrafted
assembly managed to deliver approximately 1 megaFLOPS. A simpler, albeit much slower and less expensive version, implemented using a more traditional serial processor design rather than the 6600's parallel functional units, was released as the
CDC 6400, and a two-processor version of the 6400 is called the
CDC 6500. A
FORTRAN compiler, known as MNF (Minnesota FORTRAN), was developed by Lawrence A. Liddiard and E. James Mundstock at the
University of Minnesota for the 6600. After the delivery of the 6600 IBM took notice of this new company. In 1965 IBM started an effort to build a machine that would be faster than the 6600, the
ACS-1. Two hundred people were gathered on the
U.S. West Coast to work on the project, away from corporate prodding, in an attempt to mirror Cray's off-site lab. The project produced interesting computer architecture and technology, but it was not compatible with IBM's hugely successful
System/360 line of computers. The engineers were directed to make it 360-compatible, but that compromised its performance. The ACS was canceled in 1969, without ever being produced for customers. Many of the engineers left the company, leading to a brain-drain in IBM's high-performance departments. In the meantime, IBM announced a new System/360 model, the Model 92, which would be just as fast as CDC's 6600. Although this machine did not exist, sales of the 6600 dropped drastically while people waited for the release of the mythical Model 92. Norris did not take this tactic, dubbed as
fear, uncertainty and doubt (FUD), lying down, and in an extensive
antitrust lawsuit launched against IBM a year later, he eventually won a settlement valued at $80 million. As part of the settlement, he picked up IBM's subsidiary,
Service Bureau Corporation (SBC), which ran computer processing for other corporations on its own computers. SBC fitted nicely into CDC's existing service bureau offerings. During the designing of the 6600, CDC had set up
Project SPIN to supply the system with a high speed
hard disk memory system. At the time it was unclear if disks would replace magnetic
memory drums, or whether fixed or removable disks would become the more prevalent. SPIN explored all of these approaches, and eventually delivered a 28" diameter fixed disk and a smaller multi-platter 14" removable disk-pack system. Over time, the hard disk business pioneered in SPIN became a major product line. == CDC 7600 and 8600 ==