The IBM 604 Electronic Calculating Punch

Photo
(Photo: IBM)

The IBM 604 Electronic Calculating Punch with Type 521 Card Reader/Punch, 1948. The 604 performed addition, subtraction, multiplication, and division hundreds of times faster than any of IBM's earlier electromechanical machines, and was the first IBM product to use modular vacuum-tube based pluggable units, later used in IBM's NORC and 701 computers. The 604 was programmable via plugboard and could execute a program of up to about 60 steps. Footprint: 53 by 33 inches; contained 1100 vacuum tubes and 125 relays. Power consumption 7.59 Kva. Weight: 1949 pounds. More than 5000 were sold (or, rather, rented at $645 per month, 1948 dollars, for the 604 and 521).

Watson Lab's Herb Grosch recalls [57]:

The final occupant of [the "waist" of the new IBM Watson Scientific Computing Laboratory at Columbia University, 612 West 116th Street] was the prototype, good old Serial Number Zero, of the enormously successful IBM 604 [which I installed in 1949, and it was still working away fourteen hours a day when I left in 1951]. When Mr. [Thomas J.] Watson decreed the SSEC, he also set wheels in motion "to use these electronic capabilities in the IBM." The circuits had been patented before the war, mainly by one Halsey Dickinson, whose photo showed up in the SSEC brochure.
. . .
The punch unit was much more sophisticated than the one on the 603, and the electronics box was five feet high and quite handsome, at least to someone who had had a 285, or racks of skip bars! On one end of the electronic unit was a one-panel plugboard, and it used ordinary plugwires - none of that ENIAC coax. The thing that fascinated the early users was that the machine was what we would now call microprogrammable; you could put together counters of various sizes, perform a completely flexible string of commands pipeline style, carry data over from card to card - and you could divide!

In those days one of the commonest problem areas was matrix arithmetic: sets of simultaneous linear equations, and so on. I had had to do several sets of sixth order, to pretty high accuracy, in my thesis work. On the Marchant that took most of a day, and the time went up as the cube of the order - that is, a set of twelve equations would have taken eight times as long; say a full week. At the Watson Lab this was a common exercise, involving a fairly complex pattern of eight plugboards and careful sorting. On the 601 [Multiplying Punch] the time dropped by a factor of about five compared to a desk calculator, which meant that if we had been paying rent for the machines and the building, and so on, hand work would have been about as cheap...

The 602 and 602A [Calculating Punches] helped, but not by much. But the 604 ate the problem for breakfast, or so it seemed at the time; cost per hour, if the shop had been run for pay, would have been about the same, and the speed was at least six or seven times the 602A, and nearly ten times our vanishing 601s.


More images:

604 with panels open
 
Keys and Lights
 
604/521 Control Panels

Also see:

References:

Offite Links (last checked 9 May 2003):

Most recent update: Mon Jan 12 15:36:37 2009


Frank da Cruz / fdc@columbia.edu / Columbia University Computing History