The comment thread on this post segued (oddly enough!) into a discussion of supercomputer designer Seymour Cray and a comparison of his multi-million-dollar systems with today’s ordinary personal computers. I thought it might be interesting to take a look at a supercomputer from almost 60 years ago–the Naval Ordnance Research Calculator (NORC), built by IBM for the US Navy and delivered in 1954, which held the computing speed record for several years.
NORC came only 10 years after the computer age was kicked off by the announcement of the Harvard-IBM Mark I (click here for an interesting contemporary magazine article on that innovation), but it was vastly faster and more powerful. NORC’s arithmetic was done at the rate of about 15,000 additions or 12,000 multiplications per second, and the machine could store 3600 words (16-digit decimal numbers) with a memory cycle time of 8 microseconds. Lots of NORC information and pictures at this site. Applications included hydrodynamics, weather forecasting, logistics simulations, and the motion of celestial bodies. The hydrodynamics problems included studies of torpedo cavitation and of the earth’s liquid core. (Remarks by John von Neumann at the NORC dedication, including audio, here.)
NORC’s circuits used vacuum tubes–9000 of them—and the memory was electrostatic, employing a what were basically TV picture tubes with bits stored on the face as charges and continually refreshed. This technology represented the best speed/cost tradeoff for a high-end computer at the time, but it was very sensitive–apparently, a woman wearing silk stockings walking near the computer would likely cause memory errors because of the static electricity generated. (No doubt leading to much speculation about the correlation between female hotness and computer memory error rate.)
Construction of NORC cost $2.5MM, which equates to about $20MM in 2012 dollars. Some of the cost can probably be attributed to the one-of-a-kind nature of the machine and the pull-out-all-stops-and-make-it-the-fastest spirit of its design. But even a computer intended as a standard commercial product, the roughly contemporaneous IBM 701, went for about $1 million in early 1950s money.At first glance, it seems hard to believe that such a massive investment for such relatively slow and limited machines (by our present-day standards) could have made economic sense. But consider: a calculation taking 30 minutes on NORC might have required something like 30 person-years if done by human beings using the desk calculators of the time. The economics probably did make sense if the workload was appropriate; however, I bet a fair number of these early machines served more as corporate or government-agency status symbols than as paying propositions. (As a side note, I wonder if the awe generated by early computers would have been lessened had the machines not been so physically impressive–say, if they had been about the size of a modern desktop PC?)
NORC, which was in operation through 1968, has of course been surpassed by orders of magnitude by much cheaper and more compact machines. Its computational capabilities are trivial compared with those of the computer on which you are reading this. Yet, strange as it may seem, there are a lot of problems for which today’s computer power is inadequate, and the frontiers of supercomputing continue to be pushed outwards.
While researching this post, I ran across several articles dealing with a particular highly-demanding supercomputer application currently being addressed by computer scientists. This is the modeling of the physical behavior of cloth, which is important both for creation of realistic animated movies and in the textiles/apparel industry. (See for example this paper.) Simulating the movement of a virtual actress’s dress, as she walks down the street in a light breeze, apparently requires far more computer power than did the development of America’s first hydrogen bombs.
Related post: computation and reality
