The IC3 Blue Train - A little journal about design, marketing, sales and other matters

Random Chronicles #2 – Apple Advanced Technology Group, 1987

Do supercomputers still make sense?

In 1987 (my 7^th year at Apple), I joined the Advanced Technology Group in Cupertino (CA) to manage various R&D support services. One of them was the Engineering Computer Operations, responsible for running Apple R&D’s 3000-node network and its crown jewel: a $15m Cray X-MP/48 supercomputer.

The system was running 24/7, in time sharing mode (its capacity was shared among several users), mostly for Research applications. This was rather costly. The monthly electricity bill alone topped $30k and the total operational costs were in excess of $100K per month.

In the late 1980s, $100k was about the price of a minisupercomputer, a system that was powerful enough for most applications if they could take the full system capacity. And minisupers were relatively small (one standard cabinet) and cheap to run (no need for special cooling other than air).

That’s when I formulated Aebischer’s law of supercomputing: “A full-size supercomputer like a Cray only makes sense if it is used at full capacity, as often as possible, by a single person, like a personal computer”. Otherwise you’re better off buying a personal minisuper to each of your scientific computing users. With $15m you can buy more than 100 minis and save a bundle on operational costs (about $1m per year, enough to buy 10 more minis!).

Every morning, the Cray support team published a diagram showing the system usage for the past 24 hours. I could immediately see if the monster had been used by a single user. One such application I can remember was a model for the simulation of the human ear that the researcher used to run for several hours, after midnight and until dawn, about twice a week.

Another application was the display, in hi-res color and with shadows, of a spinning top that was created from the basic physics equations (look here to see a rough simulation – the one on the Cray featured a ‘real’ spinning top with a nice wood-like texture). We used this simulation as demo for prestigious customers and visitors. Our story was that we had bought a Cray to design the next Mac and that Seymour Cray had bought a Mac to design the next Cray (he had designed the Cray 1 with pencils and paper pads). It was quite exhilarating to drive this £15m engine by the click of a mouse, I can tell you!