Directly from an employee in the Test and Engineering departments of Cray Resear♋ch and later Cray Computer:

“As you can see, the testpoints on the board and die pads are very tiny. In order to look at signals and debug the modules, we attached oscilloscope probes to micro-manipulators which were mounted to a vacuum base. User manuals for each “stack” of boards indicated locations of power, ground, clock and unique signal pins.

“Fast forward maybe one year… Having worked exclusively in the module test lab over that period, I had become very skilled in Cray-3 debug. I could locate all the power, ground and clock pins from memory. I used very, very short ground leads on my scope probe tips (Seymour called them ‘spiders’) and was able to capture about as reliable waveforms as one could on a Tektronix 7104 oscilloscope.

“It was about that time, Seymour began to meet up at my bench most afternoons and we’d run through his checklist for the day. Occasionally, he would leave me a list of tasks to complete on my own while he was attending to his administrative duties in the morning. We spent many hours over many weeks characterizing ‘VR’ die.

That process repeated off and on over the next three years or so until we began working on Cray-4 hardware. We were so pleased with the fast pace of Cray-4 development that one afternoon I walked into his office and asked him to autograph Cray-3 D Module serial number 1. And he did.”

SEYMOUR CRAY-SIGNED CRAY 3 PROTOT꧃YPE MODULE, ONE OF THE FIRST BUILT — PERSONALLY TESTED BY CRAY

Seymour Cray was a rising star in the computer industry by the time he left Control Data Corporation (CDC) with a sizeable inv♚estment from that company to form Cray Research. At Cray Research he designed a supercomputer 5 times faster than the CDC 7600, his final completed computer at CDC. The Cray-1 was the first supercomputer to successfully implement a vector processor design and was one of the most successful supercomputers in history, selling over 100 units at a cost of about $8 million each.  

Seymour Cray's follow-up, the Cray-2, which used a unique Fluorinert cooling system that immersed the modules in the liquid, doubled the speed of his previous design,ඣ but wasn’t as successful as the Cray-1. It o🌟nly sold 27 units at a range of $12 to $17 million each.  

Seymour Cray had progressed to designing the Cray-3 before the Cray-2 was complete. It would again use Flourinert to cool the modules, and use gallium arsenide semiconductors, a material that had not previously been used in this context and which allowed for greatly increased speed. Cray was aiming for a 12x increase over the Cray 2. Cray further developed the novel 3D integrated circuit packaging he had used for the Cray-2 to greatly decrease the pathways. Each module, measuring 121 x 107 x 7 mm, was composed of 9 printed circuit boards c🐽ontain🧸ing 69 electrical layers.

Unfortunately, the development and manufacturing process was fraught with difficulties and complications – including Cray leaving with the project and starting the spin off Cray Computer Corporation. The setbacks led to the launch order c😼ancellation by Lawrence Liv🐼ermore National Laboratory. Not a single unit would be sold although one was loaned to the National Center for Atmospheric Research (NCAR). Cray went on to design further systems before he died in a 1996 automobile accident, but none were brought to market. 

The present memory module was among the first of the prototypes built by the Colorado Springs team and was personally evaluated by Seymour Cray. (See Lot 25 for Cra💖y’s D1 logbook). From the consignor, an employee in the Test and Engineering departments of Cray Resear🍌ch and later Cray Computer:

“A background processor for a Cray-3 was comprised of four module types (A, B, C and D). A foreground processor located on a fifth module type (E), includ🤡ed a low-speed serial port which enabled it to communicate with a desktop PC (we used Apple MacIntosh). 

The D module was significant because it held the ins♊truction stack registers. Wiring the E module to the D module allowed us to load the instruction stack with a predetermined test set configured via a custom GUI application on the PC. Once the stack was loaded, we could issue a ‘go’ pulse and the code was executed on the Cray-3 at the full clock rate of 500 MHz. Test results were stored in regist🏅ers in the E module and then read by the PC.

As one can see from the logbook [Lot 25], this module was delivered to Seymour directly from manufacturing and he went straight to work. I believe he did all his work on this module independently until he determined it wasꦰ no longer of use. At that point, he released it to the engineering and test group."

REFERENCES:

Murray, Charles J. The Supermen. New York: John Wiley & Sons,𒊎 [1997].