Beware, this is a story that begs to be exaggerated. I was at IBM at the time but not close enough to these events for this to be a first hand telling; nor was I close enough to these circles.
The Stretch was an IBM project to build a transistorized super computer 100 times as fast as the IBM 704, a popular tube machine. It was shopped around for a while to various potential first customers, including Los Alamos and Livermore. It sounded too far off for Livermore who soon thereafter contracted with Remington Rand for the LARC. Los Alamos contracted with IBM for the first Stretch (aka IBM 7030). IBM had built no substantial transistorized machines and the Stretch was to be very substantial. A packaging of transistors had to be engineered along with the necessary design automation. This involved using Gardner Denver machines that would automatically wire wrap hundreds of thousands of wires around pins to build one such machine. Another new development was a 2 μs core memory for the machine.
The logic design of the machine pioneered overlapped execution of machine instructions, although Remington Rand was doing their own pioneering in this area for the LARC. A substantial effort was also aimed at an OS, and Fortran compiler.
The 709 was a successful product for IBM (also tubes) and there was a fairly broad customer base of users with an investment in programs for the 709—mainly previous 704 users. The Stretch was an expensive machine and not ready for mass delivery. IBM management had resisted development of a transistorized 709 for fear of cannibalizing the 709 revenue stream. (The 709s were on short term lease.) Philco had produced the Transac S-2000 which was a 48 bit transistorized computer with performance well beyond the 709.
Meanwhile, B. O. Evans had a skunk-works team of a few engineers between projects, that had access to information about the circuit and memory technology that were under development for Steve Dunwell’s Stretch project. The team knew the 709 thoroughly. (Perhaps they had come off the 709 project.) Evans knew well how high level marketing worked in IBM.
The following was internal IBM scuttlebutt: Evans sent a memo to top management recommending that IBM buy the Transac design from Philco. This was immediately forwarded to marketing to evaluate as a competitive threat. Marketing said that Transac was indeed a threat. Very soon management sent down a demand for a transistorized version of the 709.
Of course Evan’s team already had such a design, the Stretch componentry was on schedule whereas the Stretch logic design was behind schedule. Three birds had thus been killed with one stone:
The 704 customers could also move to the 7090 with little software conversion effort. The 7090 was a huge success and its revenue dwarfed that of the 709 despite its lower price. The demand curve had been underestimated! This was perhaps the first of many ‘technology translations’ that preserved customer’s software investment, not to mention IBM’s software investment. The Transac was not much heard of again.
The Stretch was late and slower than promised. Dunwell was demoted. Years later Watson realized, when someone brought it to his attention, that the 7090-7094 machines had been a major component of IBM’s financial success, and that the basic and novel technology that had enabled it was done under Dunwell’s management. Watson publicly apologized to Dunwell.
The Stretch did arrive late at Livermore but its good performance coupled with its large core memory supported their production requirements very well. It was highly checked and that was appreciated. It complemented our several 7090s very well.