Seymour Cray and the Supercomputer
Jay McCauley, 408 926 2312, email@example.com
This 30 min. tour highlights the CHM collection of "supercomputers" with emphasis on the man responsible for many of them, Seymour Cray.
What are "supercomputers"? Term from 1960-1990’s for very big/expensive machines designed to compute the hardest problems, e.g. nuclear weapons simulations, computational fluid dynamics (CFD), computational chemistry (CompChem), numerical weather prediction… These problems could not be economically solved on smaller systems, or had real-time requirements, e.g. weather prediction. Later, auto crash simulations, biological projects (genome analysis), chess (spare time…), prime number search (spare time).
1. IBM 7030 "Stretch", ‘our most successful failure.’ Scene setting for the forces driving big computers in the ‘50s (AEC, DoD, intel). Ego competition btw LASL & LLL. Contract w/ LASL (check date) 100x IBM 704/9. Transistors vs. tubes give 20x, not nearly enough. Architectural innovation [elided from most tours: pipelining, branch prediction, pre-fetch, speculative execution, write buffer…]. Alas, not enough to meet contract requirements, only 60x-80x, IBM must reduce price below cost of manufacture. Withdrawn from the market after only 9(?) units shipped. Still, team members go on to play key roles in S/360.
2. CDC 6600. Introduce Seymour Cray w/ NTDS, as one of his design signatures, small, simple PC cards. (Optional for deeper tours: Introduce Little Character & CDC 160, then the CDC 6600. 6600 has 160 CPU as a Peripheral Processor, used in a sophisticated way to fetch/store data from memory, allowing main processor to run faster.) Superscalar architecture (multiple functional units), need a good analogy… ~3x STRETCH performance. Introduce rat’s nest of wires and refrigeration units, cordwood module.
Tell story of Watson memo and the suit and pre-trial database:
Last week Control Data... announced the 6600 system. I understand that in the laboratory developing the system there were only 34 people including the janitor. Of these, 14 are engineers and 4 are programmers. Contrasting this modest effort with our vast development activities, I fail to understand why we have lost our industry leadership position by letting someone else offer the world's most powerful computer. – Thomas J. Watson, Jr.
It seems like Mr. Watson has answered his own question. – Seymour Cray
Partly in response to the Watson memo, IBM announces the 360/91 and markets it aggressively against the 6600. CDC files suit (restraint of trade?), and builds an extensive database of pre-trial disclosure materials. One of these turns out to be a memo from the 360/91 manager to the 360 Launch marketing team (parphrasing) "I understand that you intend to announce the /91 on launch day. We don’t think that’s a good idea, as we haven’t even picked the logic family for it." Smoking gun. IBM settles the suit, gives CDC a substantial amount of $ and the Service Bureau Corporation, and asks only that the DB be destroyed. IBM was also engaged in a Federal Antitrust suit, and the DB could be very harmful to their case.
3. CDC 7600, yet again the fastest machine in the world. Approximately 3-5x 6600, with maximum performance 30x. Basically the same high level architecture as the 6600, but uses pipelining to boost peak performance. Unreliable, poor up times initially.
Cray leaves CDC and founds Cray Research. CDC is an initial investor.
4. CRAY-1.Vector processor. Point at Chippewa Falls pictures in hallway as long as they are still around. New physical layout. World’s most expensive love seat, but don’t sit there, 12,000A @ 5V = 60 KW, what goes in must come out… Built using Fairchild ECL IC’s and RAM. I wish we had a card like the cordwood module.
5. CRAY-2, Fluorinert cooling. Bubbly, mad scientist cooling unit. Point out almost nobody every saw these machines except for operations staff.
6. Cray X-MP and Y-MP. First multiprocessor machines. X-MP is a Steve Chen machine, direct competitor to Cray-2, somewhat more successful.
Cray leaves Cray Research, founds Cray Computer Corp. in Boulder, CO.
[Silicon Graphics buys Cray Research, spits it out in a huge loss a few years later. Cray Research is bought by Tetra Computers and renames itself Cray Research. SGI still exists, a shadow of its former self.]
7. CCC Cray 3. GaAs implementation. Only one complete machine built, delivered to NCAR on loan. Artifact is a processor brick. 1-16 bricks in a machine (NCAR has 4). Cray’s death in car accident, 1996.
Forces drawing supercomputer era to a close:
- End of Cold War funding
- New ideas on how to attack hard problems (possible tie to Illiac 4 as prototype for massively parallel processors). Current champ "Blue Gene" at Livermore, 131,000 Power PC2 processors, same "family tree" as Deep Blue, the chess machine.
This tour has been given several times, and has been well received and is almost exactly 30 min. I’d like to keep Stretch in the tour as it sets the scene (and you walk right by it going to the 6600) and leads to the Watson memo, and Cray’s reaction, which speaks to Cray’s style. I think almost any of the docents could give this, we routinely cover the supercomputer row.
In a back of the envelope calculation, the CDC6600 is roughly as powerful a computer as an iPOD! It’s a 10 MHz clock, but gets things done in four phases, so is effectively a 40 MHz, 60 bit machine. An iPOD contains two 75 MHz 32 bit ARM processors. So easily in the same ballpark as the 6600.
I’ve avoided some of the "legends" about Cray, as I believe many of them to be apocryphal. This is probably worth a bit of research, but one has to be careful to get close to original sources, these legends take on a life of their own and get cited as if they were true.