Presper Eckert Interview



Transcript of an Interview with
J. Presper Eckert
Chief Engineer, ENIAC Computer

From: Development of the ENIAC interviews
Smithsonian Videohistory Collection
Record Unit 9537
Smithsonian Institution Archives
Washington DC

Interviewers: David Allison
Division of Information Technology & Society
National Museum of American History, Smithsonian Institution

Peter Vogt

February 2, 1988


  • ENIAC Background

    The ENIAC (Electrical Numerical Integrator and Computer), the largest and most powerful early computer, was designed to compute the paths of artillery shells, and to solve computational problems in fields such as nuclear physics, aerodynamics, and weather prediction. The U.S. Army Ordnance Department funded The Moore School for Electrical Engineering at the University of Pennsylvania to build the computer between 1943 and 1945. J. Presper Eckert and John W. Mauchly were the principle designers. The ENIAC computed a thousand times faster than any existing device. In the "Computing Gallery, Computers Before 1946," of the National Museum of American History (NMAH) on February 2, 1988, David Allison, Curator at NMAH, interviewed J. Presper Eckert about significant aspects of the design, development, and operation of the ENIAC. Specifically, the session documented both technical and non- technical aspects of the design of the ENIAC, including Eckert's engineering background, early uses of calculators to perform ballistics calculations, materials testing, and the assembly of components. Eckert demonstrated the operation of the accumulators, plug-in units, wiring conduits, and function tables with the original artifacts displayed in the gallery. Much of the session was recorded for inclusion in the "Information Age" exhibit which opened at NMAH in May, 1990. The video producer, Peter Vogt, interrupted the interview from time to time to meet script and exhibit requirements. This interview has a number of rough cuts for a professional production.


  • J. Presper Eckert Background

    J. Presper Eckert, born April 9, 1919, attended the University of Pennsylvania, where he received both a B.S. and M.S. in electrical engineering, in 1941 and 1943 respectively. He received an honorary D.Sc. from the same university in 1964. He became chief engineer at The Moore School of the University of Pennsylvania for the ENIAC in 1944 through 1946. In 1946 he became vice president for the Eckert-Mauchly Computer Corporation. He was appointed vice president for the Remington Rand Division of the Sperry Rand Corporation, 1955 - 1962, and remained in that position when the company became UNIVAC and later UNISYS. The session was recorded on Betacam master tapes, with the originals reserved for the Smithsonian Institution Archives. VHS copies and full time-coded transcripts with abstracts of important visual information are available for research use. U-Matic dubbing masters may be duplicated for presentation and exhibition. Fees are charged for copies.

    J.Presper Eckert (JPE): You know, when we were mad at this and couldn't get something to work, we used to call it "the maniac." Interface with human beings that caused us to do it this way. Some things are highly coded and some aren't. This particular thing is not coded, and the reason is that the switch is much simpler if this isn't coded. The switch only has ten connections. If that's coded, the switch would have had four resistors and four decks on the switch instead of one, so that the electrical part of that would have been four times as complicated: not the mechanical part, [it] would look the same. We would have saved some tubes on one dimension, but not on the linear selection dimension; only in the one dimension. So you might have saved--four ratio to ten--forty percent of the read-out tubes on one edge, and the overall number of tubes would have been twenty percent. By the time you got the control circuitry in there, it would have been about 10 percent, at the expense of making the switching matrix and the number of positions in it four times as complicated. It was a poor trade-off. So the whole machine was made up by evaluating various trade-offs like this, of which you picked a good example.

    David Allison (DKA): Let's go back to 1943 and the circumstances that surrounded the invention of this machine. How do you remember the environment?

    JPE: I don't know what you mean when you say how do I remember the environment.

    DKA: Was it a pressured environment? How was the problem originally presented to you?

    JPE: I had been working on a radar device during this period, which was to measure the echo time of a radar pulse very accurately. I was trying to measure something to a hundredth of a microsecond, to an accuracy of one part in 100,000-in other words, a hundredth out of a thousand microseconds-for a navigational range device that the Radiation Laboratory at M.I.T. [Massachussetts Institute of Technology] was interested in. They wished me to try to do that with analog circuits. After tinkering around a little bit with them and working in the laboratory up there for a while at M.I.T. -- where they were doing analog range circuits and things under Britton Chance, a fellow who came from University of Pennsylvania -- Dr. Chance wanted me to do it that way. I decided, kind of to his displeasure at the time, that it was a lost cause to try to do that, and we'd have to do it digitally. So I was working on a digital approach to that problem when the University stopped me and put me on this as being more important.

    Peter Vogt: Why did they stop you? What was important about it that they put you on to? Set the stage for us.

    JPE: What happened is that during World War II, a number of large guns, field pieces, were sent over to Africa. Tables were sent with them, telling them where to set the dials on the gun to allow for the amount of wind blowing and the height of the elevation of the target, with the expectation that the shell would land where you wanted it to when you followed the prescription in the book. The shells did not land where you wanted them to. The people down there using these guns were having to make guesswork corrections on the tables to hit anything, not very satisfactory in the middle of a war. The trouble was traced to the fact that the ground was more resilient in Africa, due to the type of vegetational growth in the jungle.

    DKA: It was too soft?


  • Improving military gun firing tables with computers

    JPE: Mushier or something. I don't know what term they used to describe it, but it was more resilient than the ground in Aberdeen [Proving Grounds], Maryland, where the original test shots had been made. It wasn't a good replica of Aberdeen, Maryland. So they ran some tests off down there, and by various means they found out what would happen, using phototheodolites and things to measure how shells move and so on. They found out how things were operating down there. Then the question was to recompute the tables. Their normal method of doing this was to use the differential analyzer, which, however, was too slow. I worked on that for a year with Dr. Wygant, helping to speed it up. We put 400 tubes in it, by the way, and made it part electronic. In addition, a number of people were trained to use desk calculators both at Aberdeen and at the University of Pennsylvania. Ultimately, I think, several hundred people were trained to use desk calculators at both places combined. That helped some, but none of these were getting the work done rapidly enough. In addition, new inventions were being made. Devices for airplanes to direct the shellfire from one airplane against another airplane--I think it was called the sidewise firing problem-- were being investigated, and Bell Telephone laboratories had come up with a design for such a thing which needed calculated data, and it wasn't at hand without a lot of work. So a variety of things were happening due to the stimulus of World War II, which were causing a greater need for calculated results than could be made available. IBM [International Business Machines] attempted to build some special punch card equipment, which helped somewhat at Aberdeen to do these things, under a man by the name of [Leland E.] Cunningham, who was an astronomer who had turned away from astronomy during the war to try to help with this problem in the war. Of course, astronomers are great at calculating things. So it was in this enviroment that [John William] Mauchly wrote a short write-up describing how he thought an electronic device might be built to do this: not in any great detail, just suggesting that we'd have places to hold numbers and we'd add numbers from one place and another, and we could integrate equations. He was familiar with numerical integration and numerical methods probably more than any of the other teachers in the school were. His father was a scientist, by the way.

    DKA: Did he discuss this with you before he wrote this memorandum, Pres?

    JPE: Yes. He and I used to talk about this. He took a laboratory course which I taught, which was kind of a silly thing, because the people in the laboratory in general were more educated than I was. I guess I was there primarily to point out what the equipment was for and to keep them from breaking it. There was a course with about thirty-some people in it, provided by the government during the war to try to take people that were in one branch of science and get them into the engineering branch to help with the war. It was called the ESMWT program, I believe. I think in this class of thirty people, sixteen of them were Ph.D.'s--but in other subjects. Like Dr. [Arthur] Burkes--Dr. Mauchly had a Ph.D. in physics, for example. So I hadn't gotten my master's yet; I was still working on that. But in some fill-in teaching work and in this laboratory course, I was supposed to be the teacher. But of course in engineering--I had just finished engineering school, so I was pretty well up on that stuff. He and I talked about this idea in the back-of-the-envelope stage, made little sketches on bits and hunks of paper when we'd go out to lunch together and talk about the ideas.

    Peter Vogt: What idea?

    JPE: About the idea of building an electronic computer. We first weren't sure whether it should be all electronic digital, or whether it should be partly analog. The first proposal he and I talked about was one where we would integrate pulses to do the equivalent of integration in Dr. Bush's machine, which our analyzer was the descendent.

    DKA: This is the differential analyzer.

    JPE: The differential analyzer, which our machine was the descendent of Dr. [Vannevar] Bush's machine. We were thinking of integrating signals with counters, but converting those signals to rotating shafts and doing the addition and multiplying by constants and interaction, adding and so on, by mechanical devices, the same as the differential analyzer. After thinking about this for a while, we said, "Well, this is silly. Wires are cheaper than shafts and ball bearings and gears and things, and we ought to be just sending pulses everywhere to do this." Then we further said, "This is silly because pulses sent in long strings are uneconomical of pulses." Pulses, if you want to transmit the number in millions, you have to send a million pulses by that scheme, and if you want to send a million and one, you have to send a million and one pulses. Whereas if I want to send the number of million by binary bits, it only takes twenty. So it's much more economical to put things in binary or in coded decimal. Even with coded decimal, you can represent the number with twenty-four bits. So whether you use coded decimal or binary, or even if you use straight decimal like we did in the counters of this machine here, it only takes two hundred bits; in any case, whether it's twenty or twenty-four or two hundred, it's vastly less than one million.

    DKA: So you discussed this at coffee?

    JPE: Yes, we discussed this at what might be called the back- of-the-envelope stage. We finally came to the conclusion that if you're going to do this, you ought to do it whole hog and make everything in sight digital. Then the selection of which digital system to use, whether it was binary or decimal, coded decimal, or some combination , which is what we actually used in different parts of the machine, was simply one of fitting in with the human requirements of being able to use the machine, on the one hand, and the connection to the punch card input-output equipment. [This equipment] already existed, and was the only fast input-output equipment available. There was only two choices for input- output equipment at that time, and thus, we diluted the effort by working on that. One of them was punch cards and the other was teletype tape. Teletype tape was vastly slower than punch cards, so the choice was reduced, really, to one choice. There were two kinds of punch cards, the Powers cards and the Hollerith cards. The Powers machines were interconnected mechanically and didn't fit well with electronic equipment, so that reduced it to where the only thing available were the IBM machines, which were descended from Hollerith's work.

    DKA: Do you remember when you talked about the digital nature of the calculations you were going to do? Did you have the feeling you were breaking new ground in what you were thinking about?

    JPE: Yes. We did consider one other idea that a professor that I had at Penn, Irven Travis, who later became a vice president at Burroughs, had suggested and had done some work on. He had the idea of building a machine, not as progressed as far in his thinking as John Mauchly had, but his idea was to get a lot of desk calculators together, like the Marchants and the Freeze and the Monroes.

    DKA: The calculators you were using at the . . .

    JPE: Yes. And equip them with solonoids, and then instead of the dials where the numbers read off, equip them with commutators or photoelectric devices or something to read them out, and then interconnect them electrically through some kind of switchboard panels. He gave that idea up after a short time, because the reliability of the machines are such that if you try to run them automatically in that fashion, they wear the machine out in a few months and you'd be spending all your time rebuilding them. The reliability of a desk calculator is not suitable for automatic operation; it's far from it. So he then had another idea. I guess it was his idea. It was either his idea or someone from G.E. [General Electric], where he was consulting, but someone between Travis and G.E., for whom he was a consultant, decided that maybe we could build a machine using a carrier, which, instead of being a voltage, would be a frequency modulated analog machine. I calculated the sideband problems in that, and decided that for the electronics of those days, it was too difficult. So based upon those studies, as a matter of fact, G.E. decided to build a copy of our machine, but with improved integrators. In fact, I got the idea of the photofollower to improve our integrators at the Moore School from the G.E. people. The integrators that we built actually worked better than theirs, because they didn't build servo mechanisms [as well]. They should have been better at it, but they didn't do as well.

    DKA: So you and Mauchly discussed the ideas, and then he wrote a memorandum to solve this problem. Was there a lot of pressure to solve this problem immediately?

    JPE: No, no. The memorandum was written more or less just to see if anybody else on the staff was interested. The memorandum was given to one of the men who later turned out to be the man I worked for. I don't think he was too interested in it, because the memorandum was lost and never distributed. He said later that they found it in his file or something, but anyway, nobody could find it. Someone had seen the memorandum. I think Dr. [Carl Covalt] Chambers or someone had seen it and mentioned it to Herman [Heine] Goldstine, Captain Goldstine, who was also Dr. Goldstine. He had a degree in mathematics. Goldstine saw the value of this idea and asked for the original memo. Since nobody could find it from the professor it had been given to, who was going to have it mimeographed or something. Dr. Mauchly got his secretary to retype it from her notes. They found her notes were intact, and they retyped it. We finally found both versions, and you could see slight differences. That memo was then shown to Dr. [John Grist] Brainerd and to various people around who were interested in it. In particular, Goldstine was very enthusiastic about it, and he asked the people at Aberdeen if they were interested, and they said, "Yes. Get a proposal together." So then we worked without sleep for a couple of days and ground out a proposal. On my birthday, which was April 9, 1943, when I was 24 years old, we went down there, and Brainerd and Goldstine presented the first part of the proposal to Colonel [Leslie Earl] Simon, who was head of the Ballistics Research Laboratory, at a time when Mauchly and I were still finishing the appendices for the proposal, which weren't quite finished yet, putting a few drawings in and finishing it in the next room. So we weren't in on the presentation; we were still finishing the paperwork when they made the presentation. They had a civilian consultant, whose name escapes me for the moment, who they prized highly, and he went in and looked at the idea, and told Simon, "It looks pretty good to me. Go ahead with it." The consultant's name was Dr. Dietrich, I believe.

    DKA: Were other people around that were skeptical of the proposal?

    JPE: We were told that day to go ahead. I came back and started having sockets screwed onto a chassis and making sketches, mostly writing a book of standards that we were to adhere to, tolerances and putting some parts on life test and doing various preparatory things on this. The government then proceeded to send around a series of experts who questioned all kinds of phases of what we were doing. We were able to set them straight, so to speak, or to quiet their worries. There were funny things that were done in the early days. I remember setting up a little cage with a mouse in it, because I was familiar with the fact that mice would occasionally eat the wiring in this kind of equipment. So any kind of wire we were going to use, we put in the mouse cage and kept the mouse not too well fed, and [would] see if he'd try to eat the wire before we'd use that kind of wire in the machine. Things like that. It sounds like a joke, but you know, a mouse could get in a big thing like this and do serious damage. It isn't a joke.

    Peter Vogt (Video Producer): When we were up at lunch, you were talking about money. You made an observation, a comparison between M.I.T.'s approach and the University of Pennsylvania's approach, that you were part of. It sounds like a great time to be reflecting on that. *


  • MIT and Harvard's Efforts Developing Computers

    JPE: One of the interesting things that was happening in parallel with all this, of course, is that IBM was building a machine for Harvard [University] out of tabulator parts. But it was slow. It took four-tenths of a second to add in, whereas our machine took a 5,000th of a second. In other words, our machine was approximately 2,000 times faster, basically. Actually, taking some other things into account, our machine was only about 1,000 times faster. [Laughs] But the basic adder was 2,000 times faster. IBM was off on that trail with Harvard because it fit in very closely with things they were already familiar with in what they were doing, and they had enough money to go ahead and carry out a big project like this. It was a great big machine, probably bigger than the ENIAC in size. At least I think it was about the same size. The other thing is that M.I.T. had a lot of money to spend on computing, and they decided the thing to do was take the differential analyzer and carry its performance a lot further, so they built integrators which were faster and hopefully more accurate, although I don't think they were more accurate. The main thing they did was to interconnect them through relays controlled by teletype tapes. That way that you could just prepare a tape and set the machine up without doing what you had to do with our differential analyzer, which was to take lead hammers and set screwdrivers and go to work on it like an auto mechanic for a couple of days. They felt they could set theirs up in minutes, instead of hours or days, by this relay teletype thing. They spent a lot of money on it. They spent more money on that machine by several times than we spent in developing the ENIAC. But it was a very limited machine; it would only do linear differential equations and maybe some integral equations with certain restrictions. It was generally a limited machine. It certainly wouldn't do what a digital machine will do. We perhaps had the advantage of being poorer than these other two people, than IBM and M.I.T. We had the advantage, although some might call it a disadvantage, of being poorer than IBM and M.I.T. That simply means that we had to think of clever ways to do things that wouldn't be so expensive and might hopefully even have more effect. To do that, we had to take greater chances, of course.

    DKA: What kind of chances, Pres?

    JPE: IBM didn't take much in the way of a chance. They took a lot of paraphernalia that they were highly familiar with and a lot of tabulators, and extended it into a large machine. I think it was called the Mark 1, as I recall. They did take a chance in one way. They were apparently unaware of the fact that the relays they were using were approximately one hundred times less reliable than the telephone company relays, although they were somewhat faster. But they didn't take too great a chance that it would work. Similarly, the people at M.I.T. were building the thing out of standard mechanical parts and hooking it together through standard relays and driving mechanism, setting it up from standard teletype tapes. So it was not much of a calculated risk. On the other hand, the advances that both of these machines made were not very spectacular, either.


  • ENIAC Design Risks

    JPE: We, on the other hand, had decided to take a much more greater risk. One of the reasons we needed to take a greater risk, in my opinion, we wanted to try something different, and to justify it, it involved taking a risk. It also looked like we could accomplish what they were doing with less money overall. It turned out that it took more money than we originally hoped, but that wasn't entirely our fault. We originally expected to build the machine with 5,000 tubes. As it turned out, we built it with over 18,000. But that isn't because we mis-estimated how many tubes it took to do something; it's because the government changed from wanting one function table to three. They changed from wanting ten accumulators to twenty. The original plan was just to use an accumulator by repetitive addition to do multiplication. We decided to build a separate multiplier to greatly speed up multiplication by fifty times or something. We decided to put a square-root divider, and decided to use punch card equipment on the input-output instead of teletype tape, which we originally were thinking of. We essentially built a much more versatile and elaborate device, and this shot the bill up three-and-a-half times or four times, about in proportion to the amount that it shot the machine up in size. But I have seen this happen in many other things since then, where the poorer laboratory does not necessarily get the poorer result. A guy with a lot of money goes out and buys the very best test equipment he can and, unfortunately, starts doing the very obvious things that everybody's been talking about that he would do if he had a lot of money. Sometimes those aren't the right things to do.


  • ENIAC Test Equipment

    [Eckert points to ENIAC while talking] By the way, the first thing we did when we built this machine, for example, was we needed about six oscilloscopes to let our people work. The normal oscilloscopes available at that time weren't fast enough to do this kind of work. I had, fortunately, working for M.I.T., worked on oscilloscope design just before this project, which was part of a little project to do some special tests up there. The only oscilloscope available at that time were some ones that cost about $10,000 from RCA, which would have done the job. But six oscilloscopes at $10,000 was $60,000; that would have been half of our starting budget for oscilloscopes. I couldn't see that. So having just designed an oscilloscope for M.I.T., I designed an oscilloscope and a special multi-phase pulse generator and so on. We had some technicians and people that we hired. It wasn't ready to put on the regular work, because we hadn't formulated our ideas enough, so I kept these people busy building the test equipment until we got our ideas well enough formulated. So it served two purposes: it helped train the people and kept the test equipment at a reasonable price. You see, those are the things you do because you're poor. It turned out these oscilloscopes were much better for the job than the RCA ones, and the pulse generators were better than anything you could buy at that time for the job. So being poor gave us much better test equipment.

    Peter Vogt: At lunch you made the comparison between what M.I.T. was doing--you characterized, not describing it in detail--a route they were taking and the resources that it demanded, and the route you had to take because you had fewer resources.

    JPE: The right-angle business.

    Peter Vogt: I don't understand that angle.


  • The "Right Angle" Business

    JPE: I used to think that what we were doing was taking a right-angle turn from where other people were going in this business. For example, IBM, in taking tabulator parts and simply putting a lot of them together, were marching down a straight line, just building, in effect, a sort of fancier tabulator. The M.I.T. people, in just building a more automatically controlled differential analyzer-- which was a descendent of Dr. Bush's first analyzer, which came from M.I.T., of course--were also just marching down a straight line. I call this linear, or straight, engineering. What we did was to make a right-angle turn from that and do it in an entirely different way. Actually, in the case of IBM, they not only just went straight down the line, but they missed a very important thing. The Mark One had no subroutine ability, which is an idea Mauchly suggested to me and I immediately thought was great. In the IBM Mark One Harvard machine, if you wanted to repeat a step in an integration 100 times, you had to make the program and then punch it out with changes in the numbers, 100 times along the paper tape that controlled it.

    DKA: So every step had to be repeated on the tape.


  • Origins of Computer Subroutines

    JPE: Every sequence of steps, every little routine you had, even if it was identical to the next one, except a change in the index numbers in it--or subscript numbers, if you're a mathematician--it had to be repunched, with slight differences on the tape. Whereas it's obvious that the machine should be able to have this routine and simply update the subscripts or the index numbers, and use far less equipment. Here we were forced upon it for two reasons. Not only were we poor--if we tried to build this machine with 18,000 tubes in the way that Harvard and IBM had built their machine without subroutines, this thing would have required a million tubes. That, of course, was ridiculous, both from a reliability and a cost point of view. So here we were driven by the nature of the problem into inventing a better scheme, which was the subroutine scheme.

    DKA: To allow for repetitive things.

    JPE: It allowed you to take something that you were going to do over and over again and simply tell the machine, through a special panel we had called a master programmer, to do this routine over and over until something happens. What happens? You might just tell it to do it hundred times, or you might tell it to do it until a certain number and a certain accumulator goes from positive to negative, from negative to positive, or do it until some criteria that you've developed has happened, and then stop doing it and do something else. But those clever little things like that were not in these other machines. I've been told since then that [Charles] Babbage had some thoughts like this in his work, but I had not studied Babbage at that time. I got this idea from John Mauchly. I don't think he got it from Babbage, either.

    Peter Vogt: Could you tell us again, what was the right angle that you took?


  • Innovative Computer Design

    JPE: The right angle was, I think, to go electronic, which increased our speed of operation better than 1,000 times over mechanical relays. The other thing was to incorporate the concept of subroutines into the device, which meant that a great deal of redundancy found in these other approaches was eliminated. After all, redundancy existed in both of these other approaches. It existed in the Harvard machine in that the instructions were just repeated over and over again, along with the tape. Those tapes were being prepared by Grace [Murray] Hopper, by the way. In the analyzer, there was a redundancy there again. You had a machine with ten integrators in it, all alike. I think the original one had six, and we finally got up to fourteen on ours, on the order of a dozen integrators. You could have used one integrator if you had some method of memory and could have simply moved that one integrator around to different parts of the problems. So we realized that digital things have a freedom not easily found in analog things, and that they can provide memory either through relays--they used Lake counters in the IBM machine, which is a little wheel-tabulating counter--either through those devices or, in the case of the electronic devices, through some kind of memory devices, which started out to be flip-flops in the early machine, except for constants. In constants, it turned out to be banks of switches, or where you put cables. One of the questions we were frequently asked is, "Why did you use so many cables?" Here again it's because we were poor. If you want to make a connection between, say, ten spots over here and ten spots over here, you want to do this with cables. What you need is twenty sockets, ten sockets here and ten here, and you need a cable. Plug one of them in here and one of them in here, and that does it. If you want to do this by switches, you've got to switch one switch, like the little ones on the board over there, to go to ten positions, and then have each of it switch to ten positions. So it takes you eleven switches. Not only does it take eleven switches, but those eleven switches are hooked to over one hundred terminals. So the amount of wiring and the amount of paraphernalia to do things in switches is much worse than if you do it by cables and plugs. That's why telephone exchanges contained cables and plugs and not switches in those days, in a PBX. Even in the main exchange, you saw lots of cables, because it's the same problem. As time went on and when we discovered ways through mercury tanks and magnetic cores and now through chips, to making memory cheaper, all these ideas fall apart. The motivations are completely different nowadays, so the means, of course, have changed for that reason. The technology has a great effect on the configurations that are used.

    DKA: Because you have a different set of problems to solve.


  • Origins of Computer Storage Devices

    JPE: Yes. It has to do with the technology. The latest semiconductor advances may so change things that our technology will change completely, and it may change our logical designs as a result. The logical designs and the technology are not things that are things apart; they do not grow apart. They are intimately entwined with each other. The reason in a modern machine is that anything you can do in software, you can also do in hardware, or you can do in some intermediate ground called firmware, or by microsteps or by whatever. It's a problem of the designer to find out what blend of those things gives the best overall mix. We had a terrible problem of the same type here, which is what caused me to think of the internal programming idea. I thought of this idea and proposed it long before [John] Von Neumann saw the stuff, by the way. The reason I thought of this is that I said, "If we ever build another one of these machines (and we don't just have the ballistics problems and the internal ballistics problem and a few specialized problems that we knew about to do), how do we decide how many plugs to put over here for programming, how many counters or flip-flops or something to put in the memory, how many switches to put over here with resistors? Because that's going to be different for each class of problems." There are some narrow classes where it's pretty much the same, and those are the ones we were catering to in this machine. But how do I do it when the problems are wildly different? I said, "I've got to find a way of generalizing this. Everything in science that ever amounted to anything was because somebody figured out a way of taking some specific problem and generalizing it." So that what I said was that we need a memory device that's cheap enough that we can use the memory device for everything, at least at one speed level. We may need punch cards or tapes or something at some other speed level, but at least at the main speed level of the machine, we need a memory device which can hold instructions, which can hold constants, which can hold variables, which can hold information on what has to happen next, and so on. The first thing I proposed for this was some disks spinning on a wheel with magnetized edges, which was an idea that had been proposed in a master's thesis by a fellow by the name of Perry Crawford at M.I.T., in connection with a fire control directive. He was working on a master's. It was a device that was never built, and my idea extrapolated [from it]. He was only holding constants or something on it, but it was extrapolated from that. Also there was a man by the name of [Harrison Edward] Farnsworth-- not the [Philo Taylor] Farnsworth that developed television, but a worker at Bell Labs--that had put voice on spinning disks for announcing temperatures or something on the telephone back in those days. So I knew about the work of Farnsworth, and I knew about the work of Perry Crawford. I wrote a memo. This is pre-Von Neumann, by the way, before Von Neumann came to visit us. I wrote a memo saying how we could put this information all on spinning disks with magnetic edges. Then I decided spinning disks with magnetic edges are really too slow to match the speed of electronics. I had invented a mercury tank device for timing purposes and for some other purposes in radar, an acoustic device in which you can store information for a limited period of time. But it dies out after a millisecond, or whatever the length of the tank determines. I thought of repeating that. It was later suggested that Dr. [John Vincent] Atanasoff also had that idea in connection with some capacitor memory that he had on a spinning drum, but I don't think he ever got it to work to start with, and in addition, I didn't hear about the idea at the time. I thought of this one.

    DKA: So what's the idea?


  • Mercury Tank Memory

    JPE: The idea is if you had a mercury tank, let's say you can put 1,000 pulses into one end of the mercury tank before any starts spilling out of the other, this is a memory, but the trouble is, after, say, a millisecond, if the pulses are going in at one every microsecond, it's gone. But supposing I take the pulses coming out and reshape it and put it in every time. Now the thing will sit there and recirculate, as the saying goes.

    DKA: So the storage is storing by motion of the mercury?

    JPE: It's stored by a wave going through the mercury. The mercury's standing still, but a compression wave is going through it. Particle motion is occurring back and forth, and a wave is transmitting through it. How did I think of this idea? I remembered that when I was a little boy, when I went to the store, my mother would tell me, "I want you to get these four or five things." Rather than write them down, I did as other little boys probably did when their mother sent them to the store, I repeated these five things to myself over and over again all the way to the store. In this way, my short-term memory at a young age was turned into a long-term memory to get to the store. That's the same principle. It's taking this stuff out of the mercury tank and recirculating it through a recirculating path. We built the first UNIVACs with that type of memory, the UNIVAC I.

    DKA: This is after the ENIAC.

    JPE: After the ENIAC. We built a BINAC before that, as an experimental machine using that type of memory. I was going to build a machine called EDVAC at the University of Pennsylvania, but due to a hassle that developed at the University of Pennsylvania, I left. Therefore, other people tried to build the machine that I didn't build, and they managed to foul it up and it never got really built decently.

    DKA: Pres, lets go back. Describe to me what your job was in building this machine. What was your responsibility?

    JPE: I was the chief engineer.

    DKA: What did that mean?

    JPE: That means I was responsible for everything done on the project. The only thing I was not responsible for was, I did not hire--the majority of the people were hired by Dr. Brainerd. Many of these people were students that were there or came around looking for something to do during the war that would be useful to the war effort, or had contacted the University. He was, of course, a full professor at that time, and he was the person in charge of these special projects. He was in charge of this project and about six other projects. He was the guy that brought these people in and decided what project they would work on. Ours was the largest of those projects. There were a few of them there that were friends of mine that I got him to hire, but in most cases, they were people that were already studying there or had come to the University for some reason or another. One of them who worked on the project, not in the beginning, but after it had been running for part of the year, was Tuan Chu. Tuan Chu had come from China, where his father was president of the largest railroad in China before the communists grabbed it from his father. He had come to the University of Pennsylvania to teach oriental languages. He had gotten interested in engineering, had taken electrical engineering, and when he got done, got into this. So people accumulated from various odd backgrounds.

    DKA: So you had a number of people working for you, but your job was to tell them . . .

    JPE: There were twelve professional people in toto working on the project. Several of them, like Dr. Burkes and Dr. Mauchly, had Ph.D.s.

    JPE: Leaving aside who hired the people and matters of that type, my responsibility was to carry out Dr. Mauchly's proposal. It was, to some extent, my proposal, too, because I had worked on the proposal with Dr. Mauchly. Dr. Mauchly more or less proposed the overall idea and suggested the idea of subroutines and things like that, and I proposed methods of carrying out some of these functions with counters and this, that, and the other, although he had suggested counters as one possibility, too. But when we very first started the thing, we did not know exactly how we were going to carry out the programming. So in carrying out my overall task, I had to follow up all the leads. We had sessions in which we decided on standards to which things would be built so the reliability would be great enough for these thousands of tubes. But I remember we had started into the machine and had counters that worked, and had still not decided on a method of programming how these counters would interact with each other exactly. Dr. Mauchly at this point succeeded in getting the flu and was home in bed. I remember spending one afternoon at his house, saying, "I'm sorry to come and bother you, John, in the middle of your having the flu, and I know you feel awful, but I cannot leave it go any longer than today to decide how we're going to program this thing. What do you think we should do about the programming?" He said, "I visualize some boxes, where a box would receive a stimulus here and it would have a switch or something on it and would cause the thing to happen. Then maybe it would give out a pulse to another box." I said, "I had something in mind, too. We'd have to put a flip-flop in this box that remembered which one was happening."

    DKA: What's a flip-flop?


  • "Flip Flop" Memory

    JPE: A flip-flop is a binary memory for one bit of information. I said, "We would have to put this, that, and the other together." So he and I, while he was sick and could just about see what he was doing, evolved a method which we actually used on this machine to do the programming of the sequences. Then we decided how we would have extra counters, called the master programmer, to reroute things and so on, to allow subroutines to take place, which he and I both thought were very important. It was originally his idea, but the minute I heard it, I said, "Boy, that's for me."

    DKA: Can you simplify that, Pres, the way it worked? In simple terms, what was the programming method?


  • Early Programming Methods

    JPE: In an accumulator, you had a bank of counting circuits, like this thing here. First you had to tell whether this thing was supposed to put out a number or receive a number. Then you had to tell, when it received the number, "Should it add it to the number and itself, or should it subtract it?" You also had to say, "Where does the number come from?" Then later on when you wanted to send the number somewhere, "Where does it go to?" So we had boxes on the machine- -actually, as I recall, we had some boxes which were called transceivers, which could receive a pulse, and then later when that function, which was determined by switches connected to that box, was finished, would send out a pulse to another similar box. We had other devices called receivers, which would only receive information. One can show by analyzing a set of boxes that are interconnected like this, that there needs to be more receivers than transceivers. So that's why we had a combination of the two. In the chassis that plugged in, a box would contain either one transceiver or two receivers. So these plug-in chassis, which were labeled transceivers and receivers, were connected to switches, and they were connected to little plugs on the front panel of the equipment like this. Cables, which I described before, similar to a telephone exchange, were used to interconnect them. The one special situation is that everything we did wasn't just a straight series of events like the Mark One at Harvard's tape. But after we went through a dozen or one hundred or some number of steps, we would then want to maybe redo that whole thing with some different numbers. We would maybe want to combine it with some other steps in a different way. We did that by routing it through one special panel called the master programmer, which had various counters and facilities for rerouting things under various criteria.

    DKA: What were your big problems?

    JPE: Probably the thing that was most time consuming was determining how to go about this to get reliability. To do this, I talked to people in the telephone company who had experience with tubes that were used to operate the Atlantic cable, where it cost tens of thousands of dollars, or more than that, I guess, to send a ship out and haul up the cable and change a tube. In later ones, they put several amplifiers with automatic switching between them to cut this problem down. Part of it, then, was to talk to people in the telephone company and see what they knew about tube reliability. I found out that they built special tubes for this and didn't know too much about it if you used standard tubes. I then went to RCA [Radio Corporation of America]. The head of the research department at RCA--who had offered me a job earlier, but I declined it to do this work instead--he put me with some of this Ph.D.'s and very advanced people on tube design. RCA was probably the most advanced tube development company in the world at that time. They said they didn't have an awful lot of experience at the kind of high reliability I needed, either, but that they had built some instruments in which they had run the filaments of the tubes a little lower than normal voltage and used them at less voltage and current than they were normally used at. That was done, incidentally, for other reasons; it was done to get low input current into an amplifier. But they observed when they did this that they got greatly extended lives in these tubes. So I took a cue from the experience that they had with these special amplifiers they had built at RCA, and dreamed up a set of rules that I thought would work. We then put some tubes on test and ran them in a couple of different conditions and gathered a little statistics, and based our design on that.

    DKA: Do you remember the rules?

    JPE: We ran the tubes at something more than five percent, but less than ten percent less than the standard voltage that they were intended for. I think we ran them at 5.8 volts. They were intended for 6.3 or something. I don't recall the exact numbers. We never put more voltage on a tube than half of what it was rated to stand, and we never put more than a quarter of the current through the tube that it was supposed to stand. I don't say rated to stand, because the tube manual gave very inconsistent ratings. Some tubes were rated by use, which might be a fraction of what they could stand, and some were rated by what they could stand. But we found out from RCA what the tubes would really stand, and then we stuck to within about a quarter of that. This was to increase the emission life. I think on both of these we went farther than necessary, but we were in a dangerous game, you might say, and we were like the guy first building a suspension bridge. It had to stay up. We couldn't have it fall. So we were really probably ultra- conservative in how we used the tubes.

    DKA: So you were working on a risky problem, but you were being conservative in your use of tubes, and that's how you get reliability?

    JPE: Essentially I like to think of the project as a situation where we carried out a very radical idea in a very conservative fashion. So we were a strange mixture of radicals and conservatives. I think that for this type of work, that's the position that was needed. We also looked into capacitors and into resistors. Fortunately, the dean owned a third of a company that he had started, that was the biggest resistor manufacturer at that time. They were also in Philadelphia--International Resistance Company. We were able to get a lot of resistors from them, and I would talk to their chief engineer at great lengths and found out all their experience on how you get long resistor life. On capacitors, I found that most of the capacitors that we used in the circuits themselves were mica capacitors. I found that they were extremely reliable, except that some people I knew at Lees and Northrop who worked on standard capacitors for the [United States] Bureau of Standards, said that you sometimes had trouble with loose plates. I ran tests on this and found that indeed you do have this trouble, and I found some of our counters, after we built them, wouldn't work, and it was due to loose plates inside the mica. So we built a machine that would test all the capacitors before we built our equipment, and tried different frequencies on them to make sure they didn't have any loose mica plates inside the molded device.

    DKA: Sounds like you did a lot of testing.


  • Problems Building Eniac

    JPE: Yes. You asked me originally what were the biggest problems, and I'd say the biggest problems were making sure that this thing would work when we were done. By the way, the Wright brothers faced a similar problem. The Wright brothers spent some two years building a wind tunnel and trying out their wing shapes before they built their contraption. I think we're similarly careful in that respect. Other people really had built things with wings on it, that flew some way or another, before the Wright brothers, but they were more successful because of the same studied approach.

    DKA: Were there people that wanted you to get on with it, instead of doing all this testing, or did everybody agree with your approach?


  • The Studied Approach versus Cut and Try

    JPE: No, I think people were willing to go along with the testing. In the first place, you have to realize that one of our advisors was Dr. Chambers, whose father was a statistician. I'd been raised by him, in courses I had with him and some work that I did in helping him on some consulting work part-time, not part of my studies. I learned that it's worth spending a lot of time in preparation and making sure things are right, both in calculation and in tests first, and that it pays off in the long run. People were continuously amazed when we put stuff together and it worked almost the first time, almost without change, which in electronics in those days was unheard of. If you went over at RCA at those times and watched the regular electronic engineers-- Carl took me over one time to show me--doing stuff, they were sitting there with designs for radios and preliminary things for television sets, with a lot of adjustable resistors and things hooked to those things, working these things around, trying to get this thing to work almost by hit or miss. It would be stretching it to say that the work in electronics in those days was Edisonian, because it wasn't as Edisonian as some of the studies that Edison made on lamp filaments and things, but it was too much of the Edisonian approach in electronics in those days. For a subject that's as easily, straightforwardly calculated as electronics, there was much too much cut and try. Now, cut and try came from the history of mechanical engineering, where things are harder to measure and harder to calculate. A lot of historical results in mechanical things did come from cut and try. Some of this spilled over into early electronics and didn't do it any good for a while. I think we were among the first people to really pull out of this in a big way. I don't mean to say there weren't other people who didn't do a lot of careful calculations on things before us; there were. But they were in little isolated spots, not in a big full- scale project.

    DKA: So you did a lot of calculation, in addition to testing?


  • Worst-Worst Case Testing

    JPE: Oh, yes. All our circuits were tested for what we called the worst-worst case. In other words, you assume that all the parts had drifted to the worst case they could be and the circuits should still work. You have to also figure out with some logic and some calculations what the worst case is. It isn't immediately obvious as this resistor should be as high as it could be, and this capacitor should be as low as it could be, and this resistor over here should be as low as it should be, this voltage should be high, and this voltage low. There's some combination of events which is the worst thing that will happen to you. Will it work under that combination of events? Then what is the probability that all those things will happen at once? Is it something that you should worry about, or should you relax a little? A regular radio, by the way, at that time, or preliminary experimental television sets, would not work on a worst-worst case basis. If I put everything off on a worst case in a radio at that time, it would not work. So radios had a small enough number of parts in them that they worked by setting aside those that didn't work as they came down the production line. Some guy would come along with a meter and find out which part needed to be changed to make it work. Electronics being built that way was not being built on a worst-worst case analysis.

    JPE: The other aspect was to make sure that everybody had an assigned task that fitted in well with the time scale that we were facing and that meshed in with everything else. That took a lot of my time. The other thing was to make sure that the logical designs in these things were carried out properly, so that the various panels in the machine would mesh with each other logically. The most difficult part of all that is not doing it, so much as it is calling a halt to having too much work done on some of these things. You have to have a point where you say, "I don't know. We've reached a point where this will work. It meets our goals for the moment. Sure, we could go on working on this for a lot longer and probably do better, but we can't afford to. This is the point at which this design is frozen and we're going on to the next problem now." So I had to be the nasty man who said, at some point, "This is it. We are not going any farther on this problem. We are going now to the next problem. If we don't, we'll never get the thing done."

    DKA: How did you make those decisions, Pres?

    JPE: Not by any mathematical means; by intuition. By just deciding when the thing was good enough and that time was now more important than getting a better solution to a particular stage of the problem.

    DKA: Were you under a lot of time pressure?


  • How the Name of the ENIAC came about

    JPE: It took us longer to finish the machine than I had hoped, but as I told you, the machine turned out to be three-and-a-half times as comprehensive as I'd hoped. We'd originally hoped to build this thing in a year and a half, and it took us two and a half, roughly. Part of that was the difficulties we ran into, and part of that was just the plain size of the machine. The reason the size of the machine grew was [that] originally it was supposed to be a numerical integrator to solve just ballistic trajectory and nothing else. Colonel Gillen, who was the contracting officer in the Pentagon, in his wisdom, when he named this thing, called it electronic numerical integrator and computer. He said he put the word "and computer" in there because he knew we might want to make extensions to this, and he didn't want the General Accounting Office to say, "Why did you allow extensions on this thing?" So if you put it in the name that it's going to include more than the base things to start with, you avoid that problem politically, if you understand. So in his wisdom, he did that. It turned out that it happened this way. The Ballistics Laboratory said, "We're not only interested in what happens to a shell after it's come out of the big rifle or gun, but we want to know how it's doing as it's traveling down the barrel," which is called the internal ballistics problem. "And we have some Monte Carlo problems involved in our work, and we have some other problems that we want to look at." So some of those we were able to do, and some of those we couldn't do, but we tried to amend the machine in such a way as to take on these additional requests for ability as time went on. Herman Goldstine worked on a lot of those ideas, too, and Mauchly. We all worked on those, what was reasonable to go toward in the way of additions. So did Dr. Burkes; Arthur Burkes worked on a lot of those things. No one person was responsible for all those, although it was my problem to see that they were all done.

    Peter Vogt: Was there a time when it became clearly apparent that this thing was working?

    JPE: Yes, that occurred one year after we started.

    Peter Vogt: Could you back up and give us a little sense of unsureness or apprehension about whether it would work? In other words, you set off on a course. Were you confident about the course you set out on? Share with us your feelings or anxieties about it. How did it work out?

    JPE: People have asked me sometimes whether I was worried about whether this thing would work. I was never personally worried, because I felt that we could always take enough precautions to overcome any obstacle. However, I don't know whether other people were convinced until about one year into the project. At the end of one year, we had built two accumulators and an experimental power supply, and hooked the thing up. We were able to do some very simple equations which Dr. Mauchly--with two accumulators, you, for example, can calculate a sine and a cosine wave with a simple difference equation. You can also calculate an exponential with a simple difference equation. You can do several rather elementary difference equations with only two adding machines.

    DKA: So this was one-twentieth of the machine, basically?

    JPE: Yes, you're right. One-twentieth of the machine, but it was only one style of panel. The final machine had in it fifteen or eighteen, or whatever it was, different styles of panels, and it was only one style. But it was also determined that the other panels, with one or two exceptions, would contain very similar gear. Those exceptions are that there were three panels full of telephone relays which the telephone company had splendid statistical information on how reliable these were. They were outstandingly reliable, and so we knew those panels would work if we wired them up properly. We found out exactly what the techniques for wiring were to do this, and we hired telephone people to work in their spare time on helping us do it.

    DKA: I've gotten you off the track. You have a year in the project, you've got two panels and a power supply.

    JPE: The other panels we had to design were these function tables, and these involved putting voltages in at one end, of some 600-and-some volts signal and getting out a voltage of only a couple of volts at the other end, and detecting it reliably on some special tubes that we used, some special television style tubes at the other end. That was certainly different in these two panels. But if you leave out the relays and the function table panels, the rest of the machine was essentially contained in its feel and spirit and everything in those first two panels. So once we had done that, we had a very major part of the thing proven it could be done. We did discover something in that first panel that caused trouble. The primary thing we discovered is that tubes, when they're wired up, chassied like this, even though they're apparently all alike, will sometimes break into parasitic oscillations locally due to just slight differences in the wiring or tubes from one chassis to the other. This can be prevented by putting what were called stopper resistors, or parasitic supressor repressors, in series with the grids of all the switching tubes other than those in the counters, which are controlled in. . .

    Peter Vogt: How did it look as you set out, in terms of the task to accomplish, and when did it become obvious that you were going to make it?

    JPE: After one year, and we'd built these two accumulators and overcome certain problems with parasitic oscillations, once that was done, I was sure we were in business.

    DKA: Did you run a test? Did you have the brass in?

    JPE: We brought some of the people from Aberdeen in at that time, yes.

    DKA: Then you ran a test for them?

    JPE: We had earlier experts in from Aberdeen, who had questions like, "How are you going to get the heat out of the room?" and various other things which are easily solved problems, and some guy that didn't understand saturation in amplifiers, and thought the whole thing would be unstable. All kinds of strange people came in and made all kinds of strange objections, which we were able to set to rest one way and the other. But none of that bothered me; that was just training the unsophisticated or something. What bothered me is just what kind of reliability we'd get when all this stuff was put together. I haven't touched on it, the problem of interaction of all these circuits. One has to worry about where he puts bypass capacitors and whether the number of them needed is a reasonable number. So we did a lot of work on the bypassing problem and set up some rules for doing that. Once we did most of that on those first two panels and learned that, we didn't have too much trouble from then on. Our troubles from then on were just getting it done, the complexity of it, and also getting supplies through this whole period was a terrific job. Herman Goldstine managed to get us a lot of parts that were ordered for the Signal Corps that turned out to be for machines that didn't break down often, so we were able to get a lot of replacement parts from them. I originally planned to design this machine with only four types of tubes, but I couldn't get enough of those four types, so in some circuits I had to use other types that were available. We ended up using ten types instead of four for that reason. However, we took advantage of that. When we couldn't get the right tube, we would find another tube which in some special part of the machine did this job better than the one we were looking for, but only in that part. So then we rearranged things slightly and took advantage of it. So it was a better machine by using ten types of tubes, but I would have preferred to do it with four and kept it a little simpler.

    DKA: When the Army asked you to increase the capability of the machine, was that something that you agreed with, or was it a problem for you?

    JPE: No, I think we wanted to make it a more general-purpose machine than it started out to be, and we were delighted to do it.

    Peter Vogt: Let's talk about that a little bit, about what you wanted to do and about how the Army's wishes fell into your plan.


  • Calculations That Could Not be Done Without a Computer

    JPE: We wanted to build machines which--if you had ever just been through an engineering course, which I had been as an undergraduate and had some graduate work--you realized that for several generations, people had known how to compute a lot of engineering stuff which nobody, in fact, was computing because it was too difficult to compute. It took too long. Some of these calculations were practical for astronomers, who sometimes were willing to spend ten years or fifteen years calculating how something moved out in space because the whole time scale out there is a pretty long one. As engineers, we were not afforded the luxury of time to compute that an astronomer has. So we wanted to try to carry out some of the kinds of incremental finite difference calculations and so on that an astronomer does, integrations and things, integral equation problems. We wanted to carry them out in a reasonable length of time. We had been frustrated, both John, as a physicist, and myself as an engineer and student, in finding that after you'd learned all the theory, you couldn't apply most of it because of the calculation problems. I always think of it as like having a nice larder or a pantry full of canned goods, and you don't have a can opener. Here's all this good stuff up there; you can't get into it. By "good stuff," I mean for several generations we'd had great equations for designing filter circuits and various other kinds of circuits in electrical engineering and in a lot of things in mechanical devices. For example, I could write the integral equation for a transmission line, and I remember I spent a lot of time looking in every language for solutions to that problem, and never found a satisfactory solution. I found one in a French article that had a series in it, but didn't converge rapidly enough to be useful. So actually, the telephone company built models with coils and condensors, of their tranmission lines to solve the problems in those days.

    DKA: So you were interested not in solving the ballistics problem so much as something else.

    JPE: Engineering problems, scientific problems in general, of which the ballistics problem was the convenient problem for which [as it] would have been called in show business there was an "available angel" to finance it.

    DKA: You knew that even as you were designing it?

    JPE: Oh, sure. As a matter of fact, shooting at people wasn't our bag.

    DKA: What was your motivation?


  • Motivation for Building the Eniac

    JPE: Our motivation was to build a machine which could take all these great mathematical ideas, which mathematicians and engineers and scientists had dreamed up for generations before we came along, and be able to exploit them and use these great ideas in a reasonable length of time. Sure, most of these ideas could be done if you had years and years to do the calculations in. But engineers don't have years and years to do the calculations in, and neither do scientists, in many cases, have years and years to do calculations in, astronomers being perhaps the one exception up to that point in time.

    DKA: So the war gave you an opportunity?

    JPE: It gave us an opportunity for someone who was interested in doing a problem which fit something we would like to do.

    DKA: And it gave you money and supplies.

    JPE: Money and interest from people to do it. I personally was kind of glad that I didn't have to be out there shooting at somebody, because I don't think that's a very profitable thing for society. I was really delighted to be working on something which I thought would have uses other than military uses when the war was over, and which I thought the uses would be more important than the war once it was over. The war is important. Of course, you can't be killed in the meanwhile; you have to survive. But once having survived, I thought that this device would survive and be more important than its original purpose by far, which did turn out to be true. We were evangelistic. We were trying to sell everybody we could get our hands on, what a great idea this was.

    DKA: And did people buy?

    JPE: Some did, some didn't in those days. When we first tried to sell businessmen on the idea of using magnetic tape to store their information, one of them, I remember, said, "Well, suppose my competitor gets a big magnet and puts it in a truck and drives by and destroys all my records." My answer to this was, "I worked on magnetic mines, and magnetic fields vary as the cube of the distance from a magnet, and the size magnet you would have to put on the large truck you can imagine is beyond getting down most of our streets to do that. It's impossible to do that."

    DKA: I'd like to start talking about the machines, the components. Pres, what are we looking at here?


  • Designing Electronics for the ENIAC

    JPE: You're looking at primarily a power supply control panel which was used to check not only the supply voltages that came into the machine, but the various voltages that were used throughout this machine. This machine contained, as I recall, eighty-eight voltages. Some people wondered why we had so many voltages. The reason is, here again, one of economics and reliability. A pack of, in some cases, fifty or seventy-five wires running around in back of the machine, connected differently to different panels, was fairly cheap. A lot of extra parts to get various voltages that you wanted locally but repeated in several thousand places are not cheap. So for example, normally in a circuit, if you had only one of them in a radio, [you] get an intermediate voltage between your supply voltage and ground by putting two resistors in and getting an intermediate voltage off of it. That, however, takes two resistors. In this machine, if we wanted an intermediate voltage, we provided an intermediate voltage from the central cabling system and from the power supply, and then we didn't need these two resistors. You say, "Well, who cares about two resistors?" We did, when that might be repeated 10,000 times in the machine. Really it was 20,000 resistors. Just putting in a wire was cheaper than putting in thousands and thousands and thousands of extra resistors.

    DKA: So you used a lot of voltages to cut down on the number of parts?

    JPE: There's another reason for doing it, and that is if you have the voltages provided from a central place, and by means of a matrix here in a switching system we had, we could check any one of these eighty-eight voltages quickly. We'd just turn the meter, and the meter was normalized so it would read one hundred percent if it had the right voltage. You wouldn't have to read any meter; you just looked to see if it read within a green band here on the meter. This meter is missing right now. If we have 1,000 points in the machine where we provide a voltage to each of those points with a pair of resistors located locally, there's no easy way of finding out whether that voltage is correct in those 1,000 places. We can't afford to bring 1,000 wires back to a panel to check it, and there's always the possibility that one of them is wrong. If that voltage at 1,000 places provided is by a wire we can check in one spot here with a metering system, we have very much less risk. Another way of saying it is that not only is a lot of extra parts an additional expense, and we have to look at that side of it, but it's very much of an additional risk. This machine had 70,000 resistors in it already. We didn't want a lot more.

    DKA: So you brought those back and allowed checking and testing through this panel?

    JPE: We eliminated a lot of resistors that would otherwise have been in the machine by having what was a colossal number of voltages. Most electronic equipment being designed in those days was designed with one, two, three, four voltages. Here we come up and design something with eighty-eight! People thought we were strange.

    DKA: It sounded complex, but it was really more simple.

    JPE: It was really simpler if you looked at the overall system. Many of these things that we made decisions on people questioned because they only looked at one aspect of the problem. You have to look at all of these things as an overall system problem, and not one aspect.

    Peter Vogt (Film Producer): So it was, in effect, more simple. Explain that.

    JPE: We made the machine more complicated in terms of the number of voltages we sent around it, because it greatly reduced the number of resistors and parts that could vary on us and cause trouble in other parts of the machine. So while, on the one hand, it appeared more complicated, in the overall picture it was really vastly simpler to do it this way. But it was very unconventional to do it this way at that time.

    DKA: Whose idea was that?

    JPE: That was my idea.

    DKA: What are these buttons for? Is that how you start ENIAC?

    JPE: It's been a long while since I've looked at this panel. This turned the power on, and this turned the machine off.

    DKA: But if you're going to start ENIAC, is that what you...

    JPE: That's what you pushed. That button was never pushed, except maybe once every month or something, because we left the heaters in this machine on all the time. It's turned on by this and this. Whenever we wanted to work on it, all we did was turn the DC voltages off and left the heaters on. The reason is that generally speaking, if we were to turn the heaters off and on once, generally speaking, we would blow a tube. The expansion and contraction inside the tube from heating and cooling it--each tube having a couple of dozen welds in it, or other things, little fine wires in it--the thermal change brought about in the dimensions of the tube by doing that would probably break one of them in 18,000 of them. If you left the heaters on and turned only the DC power off, this didn't happen. The only time we turned the heaters off was for some special reason, like a storm which disturbed our voltage source or something like that, and had to do it.

    JPE: This panel looks pretty much as I remember it, except these lights were added since I last saw it. I don't know what they're for. But normally, to start the machine, one pressed this button here, and that turned on the heaters of the machine and it also turned on the DC voltages. We could also turn the DC voltages off and have only the heaters on, which we frequently did during testing, because turning the heaters off and on, due to thermal expansion of the tubes, frequently blew a tube. In fact, if you turned the machine off and on once, you could almost count on having one blown tube. Otherwise, the machine would go for one or two days at a time without a blown tube. This button over here is the initiating switch. This is the button that gave you the first pulse that started the program into operation. Normally, we didn't do it by pressing this button. There was a button on a cord, a portable button, that could be plugged into one of the trays that carried wires around the machine. You could actually work this initiation from any where around the machine, and that's because lots of times you were testing a particular panel and didn't want to run halfway across the room, which was fifty feet long, and press this. So you plugged this thing in where you were, and pressed it there to start the machine again. The stop button simply killed the power supplies. These switches up here, this one here simply checked the three different phases of the three-phase power supply, to see that we were having the right AC voltage from the power company. As I said, these others were used to test the voltages.

    DKA: Talk about these.

    JPE: These were mostly connections from this initiating circuit so that when you pressed the button, the connections to go to various parts of the machines. Here again, some of these have been added since I've seen the machine. Some of these were not on the original machine, some of these outlets here. Some people found some modifications desirable in the machine as time went on. I don't know what they were.

    DKA: These descriptors--reader, printer, and pulse--describe what?

    JPE: Here again, by plugging this portable initiating cable into one of these, or connecting this to one of the trays and plug in anywhere, you could trigger the machine that read punch cards, you could trigger the machine which punched punch cards, which is called the printer. This was an initiating pulse from some remote point. These were connections to remote points for controlling various things remotely.

    DKA: Wouldn't you say this is one of the basic racks? Isn't this the start.


  • ENIAC - Computing Processes

    JPE: This unit here is the basic high-speed memory unit, or element. While this machine, or a remote connection to this machine here, can be used to start the process, the process of computing in the ENIAC was largely carried on by panels called accumulators, which are behind me. They comprise roughly half of the machine in terms of electronics. Each of those accumulators might be thought of as an electronic adding machine, except with two very important differences between them and a regular adding machine. One is that they were electronic and added to numbers or subtracted numbers in a five-thousandth of a second. The second difference is that they contained a program control system so that you could tell them to do an addition, and then tell somebody else when they were finished. You could also control whether it was an addition or subtraction, or you could tell it to simultaneously transmit a number to one place as a positive number, and simultaneously transmit the same number to another place as a negative number. The top half of the panel, contained the adding machine part. The bottom half of the accumulator panel contained the control sections. The top part was made of ten panels that looked like this, plus an eleventh one for the sign. This device we called a plug-in unit, and in its day it was kind of an original idea in electronic equipment, to plug in as many parts of the machine that were repetitive as possible. The first ten tubes here are a device that can count up to ten electronically at the rate of 100,000 counts a second. We designed this circuit first before anything else. We tried six different circuits out, and finally modified one of them and picked this one as the best one we could find. It had far more ability to count accurately and withstand changes in voltage and part values than any other circuit we could find. We, nevertheless, found it was critical of the shape of pulses that were fed into it, so we designed a special circuit called the pulse shaper, which was found in the next three tubes here. Going to the other end, there are six tubes down here which comprised a transmitting device for plus numbers and a transmitting device for negative numbers, which are controlled by this counter. There are a number of tubes, as you see, intervening here. I think there are twenty-eight tubes, and we've only accounted for thirteen and six, or nineteen of them. The rest of these tubes in here are primarily concerned with carry-over mechanism; in other words, with remembering when two numbers add up to more than nine. We'd have to carry it over and put it into the next channel, and to do this in such a way that the process isn't slowed down. These tubes are controlled with the switching and so on relating to the carry-over process which allows a number of these chassis to be connected together to make a full accumulator or full adding machine, depending on what you want to call it. Close-up of Eckert pointing to tubes on the plug-in unit]

    JPE: A second important part of this device are these six tubes at the end. These three were used to transmit a pulse down the line in a positive fashion, and these three are used to transmit a pulse down possibly another line, usually a different line, in a negative fashion. We could transmit numbers positively and negatively at the same time to two other places. The rest of the tubes in between these two groups, one [group] of thirteen at one end and [one group of] six at the other--these few tubes in here--were used to remember that there was a carry if two numbers added up to more than ten, and to cause this carry to be properly put in the adjacent channels and carried over from channel to channel as necessary on repeated carries. That comprised the total function of this unit.

    JPE: The wiring in this machine is not like we see today, but the printed wiring that we use today in so many items had not been invented at this time. Actually, the wiring in this machine goes back to an older era of earlier radios which were wired with solid bus bars. We used some of that technique here, combined with cabled wiring and some point-to-point wiring in a combination that made things easier and more accessible for tests than they were in, let's say, a radio built at that time. We spent a little more time planning wiring in this than one did in normal electronic equipment in those days.

    DKA: It looks incredibly orderly. Is there a planned orderliness?

    JPE: Yes. This is to make the testing easier. If you have just a rat's nest of wires running all over the place, it becomes much more difficult to test things. The support of these wires not only held things where you wanted them and offered firm places to put things to so they didn't fall around and so on, but it gave fixed points that you could show on a drawing where voltages were to be found to test easily.

    Peter Vogt: Let's talk a bit about this being the building block, foundation of accumulators. I haven't got that concept quite in my head.

    JPE: I could say one more thing about the wiring.

    DKA: Sure.

    JPE: Today, with the printed circuit wiring, this orderliness is more or less inherent in the circuit layouts, although today we are making boards with up to twenty layers of wiring, so you can only see the outer two, and eighteen of the layers may be hidden from view. So some parts of today's things are very easy and simple things. Today's stuff is much easier to look at than this, but in complicated things, today's devices are very difficult to understand inside.

    DKA: You didn't say anything about the bulbs, Pres. You might say a little bit about what they're there for.

    JPE: These little bulbs were hooked onto every flip-flop in the memory and also flip-flops that were used in any of the control circuitry or programming circuitry. So each accumulator had over one hundred bulbs in it, and since there were twenty of them, that's over two thousand bulbs. There were other bulbs on the programming circuits and other things, so there were probably over three thousand of these little bulbs around there. Each one of these bulbs was tested first to make sure it fired on the right voltage and went out on the right voltage, to make sure it was reliable. These bulbs have kind of led to a funny thing about computers, that every computer I've seen in a science fiction movie since this machine came out had flashing lights. Our modern machines do not have flashing lights anymore; they have cathode ray tubes. But most of the machines in science movies still have flashing lights, which is a throwback to this machine of some years ago.

    DKA: You had them for what reason?

    JPE: We had them because one of the conservative features about this machine is that we could rig the machine up to run it at any speed slower than its rated speed of 100,000 counts a second. We could run it, for example, one a second or even one a minute, if we wanted to. That allowed us, if we were having trouble, to slow the machine down and see exactly on which step the trouble was occurring. But we wanted to know exactly where the memory stood on every step, and these bulbs enabled us to see which flip-flops were turned on, both in the number memory and in the programming systems, so that we could do this. The alternative would have had to be to go in with a volt meter and make tests on the circuit if you didn't have these lights, and that would have been terrible. You wouldn't have been able to service it decently.

    JPE: About thirty percent of the tubes in the machine were in a chassis just like this. They were all interchangeable with one another. We had spare ones that we used to test. Thirty-percent comes from twenty-eight tubes times two hundred of these, which is 5,600, which is approximately thirty percent of 18,000.

    JPE: About thirty percent of the tubes in the machine were in a chassis just like this. There were two hundred of these chassis, twenty-eight tubes in each. That's 5,600 tubes out of 18,000, or about thirty percent. These chassis were all interchangeable with one another. We had about ten spares of this kind of chassis we kept in repair, and if we suspected trouble in any one of these chassis, we took it out of the machine and put another one like it in it. We had special handles, by the way, for reaching in and grabbing the end of this and squeezing the handle, and it would grab these by the end so we could pull them out easily without having to touch the tubes or have difficulty getting at them.

    DKA: You had twenty accumulators built with this type of unit?

    JPE: Ten of these chassis in each. In addition, each accumulator had in it one more chassis different from these, which carried the plus-minus sign controls. Then there were about eleven smaller chassis that went in the lower part of the panel, which controlled the operation of this panel from eleven possible different points and gave out signals to a number of different points. Actually, there were somewhat more than eleven, because some of the control chassis doubled up on receiving end.

    DKA: The plug wires and signals came in from the front?

    JPE: The signals all came in from connectors in the front of the panels, and went to trays between the panel and the floor that were arranged around the whole machine.

    DKA: Like those over there?

    JPE: Yes, those are trays. Actually, the trays up top generally carried digital information on the numbers, and the trays at the bottom generally, but not always, carried program information on what operation was to be performed and when. By the way, this machine was different than a lot of early machines, in that this machine could, if you wanted to plug it up properly, carry on two or three different things at once. Today it's commonplace, again, to build machines in which there are several processors working at once, but there was a period in this computer development where the machines all only did one thing at a time. So this machine, curiously enough, is like its most recent brother in that respect. Also it's like it in another interesting respect. In more recent machines, the controls for each microprocessor tend to be on the same chip with the processor, to eliminate wiring problems. Here again, the controls to control one of these panels, the programming elements, were in that panel--and you could add other panels without disturbing that, the same way you can have a machine with several microprocessors and add some without disturbing the ones that are already there. This machine has a funny connection with the present that disappeared in the interim between the present and these early days.

    DKA: Why did you use parallel processing?


  • Parallel Processing

    JPE: We thought we could get more speed by programming several things at once. We found the programming so very difficult that people rarely did it. Also, this machine didn't have enough memory to really exploit that idea fully, and it turned out to be something that we thought about and tried on a few things and exploited a few times, but most of the time it didn't turn out to be a practical way to use this first machine. But we kept the idea in the back of our mind and brought out parallel things a little after that.

    DKA: So you felt like some of your engineering ideas were ahead of what you could do?

    JPE: Yes, some of them were. Of course, this machine is parallel in another way. When it transmits information from one accumulator to another--one part of the machine to the other--it does so on eleven circuits at once, the ten number circuits and the sign circuit. In that sense, it's partially parallel. It transmitted information in a serial-parallel fashion. It was neither serial, nor was it parallel; it was a combination of the two, this machine was, in the same sense that the tabulators of those days were a combination of the two. In some respects, it was like a punch card tabulator. It was unlike it in another. The punch card tabulators did not have any convenient method of transferring from one set of registers to another. You could transfer from a punch card to a set of counters, as they called them, and you could transfer from a counter into a punch card, but there wasn't any convenient way of going from one set of counters to another. There was, however, in the Mark One. That was an innovation on the punch card art that occurred in the Mark One. Generally speaking, just the simple idea of transferring numbers from one register to another, particularly doing it electronic[ally], was original in this machine.

    DKA: Pres, you said at lunch, before we did our taping, that this machine, unlike what some people think, did not grow from a differential analyzer conceptually, but grew out of the tabulator machine. Can you say where this idea came from?


  • ENIAC Origins - Differential Analyzer or Tabulator Machine?

    JPE: There's been a good deal of misconception of how this machine originated. Even some of the professional people that worked on this machine have written articles which misstate how it originally got started. It has been stated in some articles written about this machine that we got our ideas by extending the ideas of the differential analyzer. Only in the sense that we started out to try to do the same problem that the differential analyzer was doing at that time, which was the ballistics equation, but our ideas for this machine were much more based upon ideas from mechanical desk calculators. Dr. Mauchly was very familiar with those, owns one, and had used them in teaching work at Ursinus College, and was interested in using them for statistical work, as well. I had had a course in statistics and in subjects and had learned to use one, but not as much as Mauchly had, as a student at Penn. At least I knew how much time it takes to make one work and what was involved. But he knew the numerical methods of how to integrate equations with them, which I had not learned yet at that time. I had learned a lot of mathematics in college, but most of it was theoretical mathematics, and not very much emphasis on discrete calculation methods.

    DKA: Do you want to say anything more about the calculator background and what ideas came from the calculators?

    JPE: Yes. The ideas of having successive counts in a particular column in a mechanical calculator are similar to the idea of successive counts in a counting chain in this machine here. By the way, the idea of a counter was not original with us. Counters originated with several men who developed them for physicists who were counting pulses from Geiger counters and from various energy sources involved in the study of various theoretical physics problems. So one of the reasons we were interested in the counter is that we were able to get a little literature and get a little bit of a start. I'd also built some counters for a previous project that gave me a little start in this direction. If I'd had to do it all over again, I wouldn't have started it that way, but that's the way we did start. The counter is like the adding machine. There were other things in the machine. In our multiplier, we used a partial product system, and this partial product system is done with some cam-like devices in some of the desk calculators. So that idea, which Mauchly suggested, came from his knowledge of a desk calculator. So probably the greatest influence on this machine, other than theoretical knowledge about discrete calculations, was from desk calculators and not from the differential analyzer, as has been frequently suggested. In fact, I don't know of anything other than the type of problems that we were doing, which connects it to the differential analyzer.

    DKA: Pres, you had forty different panels in the ENIAC. Tell me a little bit about how information flowed among them.

    JPE: Thirty-seven of these panels were electronic, and three of them for relays. We could best go into this by coming over and looking at the wiring troughs which interconnected the machine.

    DKA: Pres, you had forty different panels in the ENIAC. How did you get information to flow among those panels?

    JPE: Of course, three of the panels were relays, but the thirty-seven remaining panels were connected by wiring troughs which contained, generally speaking, eleven wires along the trays. Let us go over here and look at some of these.

    DKA: Pres, you had forty different panels in the ENIAC for the computation. How did information flow among them?

    JPE: Of course, three of the panels were relays, and were not involved in the main electronic travel. Thirty-seven of the panels were interconnected by wiring trays which ran in front of the machine at two levels. One set was for the digital information of the calculation, and the other was for the digital information controlling the programming of the device. Why don't we come over and look at these panels? I think we might be able to get a better idea.


  • ENIAC Wiring

    JPE: Here we have a complete accumulator showing the way it was connected to the wiring troughs, or trays, as we sometimes called them. These cables here carry information from an accumulator to a tray, or trough. These troughs ran around the whole machine. They weren't short like this one; they were long and went around the entire machine. They were about seventy feet long. They carried signals from this unit to any other unit in the machine. These two connections here allowed information to be taken out of the machine as a positive number, the other one allowed the information to be taken out as a negative number. These five here allowed information to be taken in from five other places in the machine. There were switching circuits in here that selected which you wanted to use and which way to do it, at what time. They were prompted to do this by the setting of these switches, which could be set to transmit or to receive on the various channels and in the various modes. There was also another set of wires down here that connected to a similar set of wiring troughs, or trays, that ran around the machine in a manner similar to these. These carried the programming signals which caused the machine to operate on the numbers. These were individual cables because the program signal was only a single pulse going through a single wire, whereas these cables had eleven connections in them, corresponding to the ten-digit numbers and a plus or minus sign. There were also switches that allowed certain other functions to take place on here in these little switches up here. There were also extra plugs at each end which allowed this unit to be connected to a neighbor and extended into a double precision unit, which was desired in multiplication for the product, for example, in some problems. So altogether, this provided a flexible unit. One of the things that was a little strange about this machine is that we designed this device before we knew what all the rest of the machine was going to look like, so this had to be flexible enough. . .

    JPE: Over here we have an accumulator, and up here we have the wiring conduits, or trays, as we called them. These cables connected the information for the numbers, the digits, from the accumulator to the trays. Each of these cables contained eleven wires, plus some shielding, and were used to carry the ten digits and the sign either into the trays from the machine, or out of the trays back into the machine from another machine. These trays, by the way, ran all the way around the machine most of the eighty feet that the machine was long, a little less, actually, because they weren't needed in front of the three relay panels. Down here we have another set of wiring conduits, or trays, which carry connectors that are only have a single wire in them, and these were used to carry pulses out or in, to control the programming circuits, which just required a single pulse to tell it to do a certain thing, and gave out a single pulse telling the next unit down the line to do something different. What was to be done by this panel was determined by where a switch corresponding to a plug was set here. There's a little auxiliary switch here where these were set, and they controlled the function that took place in the machine. These plugs at the end allowed the machine, by being jumpered sideways rather than the way they are now, to be connected to other panels to get double precision, such as required in the product of a multiplication, for example. At the time we designed this, we had in mind a somewhat simpler machine than we finally built, but we built this with sufficient flexibility in the way it was designed, so that as additional features, such as a multiplier and a square-root divider and extra function tables, and things were added, that we could still comply with that ability. By the way, one interesting feature is the programming scheme used here and the number of connections allowed in the trays, and the number of bus circuits allowed by these trays up here, were such that we could carry on several calculations simultaneously in different parts of the machine. We rarely did this because we didn't have enough programming ability to do that very often, and also because the programming requirements in timing when you program that way, make the programming much more difficult, and it was already difficult enough. We, nevertheless, had that ability. That ability disappeared in the machines that followed this, that we and other people designed, and has more or less reappeared but in a much more sophisticated form in some of our most recent machines. These panels--there were twenty of them--comprised half of the forty panels, and a little over half of the electronics of the machine.

    DKA: Was there any series of problems related to this kind of wiring that created special difficulties for you?

    JPE: You had to be careful. Inside these trays, for example, there aren't just loose wires; there are buses that run and are carefully spaced, and they have little metal shields coming up between them to prevent interaction of the signal. The trays used for this purpose and for this purpose are the same except for the type of connectors used out front. There would have been problems, except we carefully tested out and tried our ideas out in simple experiments before we cut it out of metal.

    JPE: When a program is stimulated by a pulse from one of these lower trays through a cable such as this one, it is communicated to the switches above it. These switches can be set to perform the operation you want to perform here, whether it's the reception of a signal on several lines or the transmission of a signal on one of two lines. As a matter of fact, a switch can be set into a compromise position in which a signal is both transmitted as a plus number on one line, and a minus number on another line. We also have another switch in here which can be set, which can cause the operation up here to be repeated from one to nine times. There are many times in mathematics when you want to multiply some numbers by a small constant to keep them from going out of range, and this allowed that to be done without tying up the multiplier or some expensive part of the machine in a manner more economical of equipment. There were a number of flexibility features about this machine that allowed it to be used with all the different parts that were later added on to the machine. In the beginning, the machine didn't have more than ten of these in mind, and was later changed to twenty. The machine, as it originally was planned, didn't have a multiplier, and we added one. It didn't have a divider or square-rooter, and we added one. It only had one of the function tables, which are banks of constants, and we added two more, having three altogether. So we had to design these things in such a way as to make it open-ended for the future, even though we didn't know quite what they were going to ask us before we were done with the project.

    JPE: These troughs, or trays, as we called them, carried the information about numbers from the machine to other machines along the long trays around the machine, or in the reverse, they carried information into this machine from other machines. Each of these plugs had twelve connections. One of them was a ground connection, and eleven of them were to carry information.

    JPE: Each of these plugs up here carried information either from or to a tray. It has twelve connections here, plus the guide pins. One of these is for ground, one of them is for the plus and minus sign, and the other ten are for the ten digits in the panel. These wires are shielded, actually, individually in these cables to prevent interaction. As can be seen up here, the accumulator had a row of ten lights for each of the ten decimal digits and had two lights for the plus and minus sign to the left here.

    JPE: Up top we see a number of small neon lights. These would glow pink in normal operation. Only one of them in a particular column would be on at a time and would indicate whether the number was zero, one, up to nine. The two on the left here indicated whether the number was plus or minus. There were also a small row of lights below that, and they indicated whether there was a carry- over digit in that column carried to the next one. Bear in mind these lights were not useful when the machine was running at full speed because they flashed on and off so rapidly that you couldn't see them, but we could run the machine one step at a time very slowly, and we could see exactly what was happening in the operation of the number under those circumstances from these little lights.

    JPE: These accumulators formed the fast memory for the ENIAC. One of these accumulators could add or subtract or transfer 5,000 numbers a second. We had another device that was part of the ENIAC, which we called the function table. That's rather a misnomer. It's really just a table of constants. The reason we call it a function table is that we had functions in the particular problems we were doing, which were called ballistic functions. This device contains switches. You are only looking at one side of it. It has a similar set of switches on the opposite side of this device you're looking at. This device has enough switches that it stores over one hundred times as many numbers as the accumulator that we just looked at. It is connected to two panels containing, each panel, about the same number of tubes as the unit we were looking at. The unit we were looking at, the accumulator, had about five hundred tubes in its back. This is connected to a device having about 1,000 tubes altogether, two panels worth. But since this thing holds one hundred times as many numbers, with only twice as many tubes, this thing holds numbers with one-fiftieth the number of tubes that the other device has. It is not quite as fast. This takes three or four times as long to operate as the other device, but the number of numbers available in the machine from the three of these units that we had was far and far in excess of what the other machine had, over three hundred times as many numbers. These switches were normally set by hand. Each of these rows contained twelve switches and two signs. Since the accumulator only had ten numbers in one sign, you might wonder why we had twelve and two here. The reason is we frequently split these up into two sixes, each with its own sign, or perhaps an eight and a four, each with its own sign. One of these tables has one hundred and four entries, but we frequently got the equivalent of two hundred and eight entries in a device like this. So the vast amount of numbers that we put into the program were set off to these switches. Other numbers were brought into the machine on punch cards. We also had a section of another panel where we could set constants like this, only they would come out at 5,000 a second, the same as this panel. But we only had, I think, twenty constants in that panel. Truthfully, the reason we had it is we had half a panel left over, and we decided to use it up that way.

    DKA: Pres, how does this indicate the nature of storage technology of the time?

    JPE: This device here today would be called a PROM, which would stand for Programmable Read-Only Memory. It means I can only read this electronically; the writing has to be done by some human process, in this case, turning switches. We now have PROMs that are operated by certain rather slow electric operations or by putting light on them in certain ways. There's various ways of doing it, which are not only 1,000 times faster than this or more, but also will hold maybe one hundred times or more what this device will hold on a little chip a quarter of an inch square. In fact, PROMs are in the works which will hold several hundred times as much as one of these devices will hold at the present time.

    DKA: What's inside this box?

    JPE: This box here that is connected to the two panels I mentioned, which have the 1,000 tubes in the two of them, this box itself is largely air inside. What is here are switches and wires. This switch connects a resistor to ten possible wires, depending on where you set it. These wires are then brought out to giant plugs out here and connected to the electronic circuits to make the overall device function. As a matter of fact, a strange thing has happened since the ENIAC was built. At the present time, the technology to build memories that can both read and write rapidly, like the accumulator did, are actually somewhat faster and cheaper than ones that will read only. But the advantage of the read-only memories is that they don't forget if you turn the power off, whereas the regular chips forget if you turn the power off. Most of the chips now in use have 250,000 bits of information, which is equivalent to maybe 60,000 dials like this on a chip a quarter of an inch square of silicon. We have now about reached the turning point in the last couple of months, where in new production, about half of the memory being turned out at the present time is in one million bit per chips. In the laboratory, people are working on four million and sixteen million bits per chip in the future. Where the limit on this sort of thing is isn't clear yet, but we are reaching a point where we're running out of what you can do with photoengraving techniques, which is what they're using now. Some people are working on techniques to make chips directly with electron beams, which are, of course, finer than light beams.

    JPE: One of the problems with this type of memory is that to come along here and set every one of these to the correct number, looking at a sheet of paper with the data you want in the memory to be taken from the sheet of paper, takes quite a lot of time, certainly like a half an hour to set up one of these panels, if only one person does it. It's also confusing. It's easy to get mixed up in the rows and columns. To help us a little, we made every fifth row with red knobs; that helps some. We made every certain number of numbers here with different color plates, to help keep track of where you are. But it still is a time-consuming and effort- taking job. Nevertheless, when this was built over forty years ago, this was a very fast, economical way of storing data, probably better than anything else that existed for a few years after that. At the present time, of course, we put data in almost exclusively from magnetic disks and not by hand.

    JPE: To set a memory like this one, you had to take a sheet of paper with the numbers on it and hold it in front of you, try to read those numbers off, and set these dials one at a time to the number, like this, going across here, until you have them all set. As you can see, this is quite a time-consuming problem, also one which is fairly easy to make errors. We use different colors on the knobs, red on every fifth row, and we made the columns different colors in a periodic pattern here to make this process a little easier, but it's still a time-consuming and not terribly satisfactory way to enter numbers in a machine. Nowadays, of course, we use the chips which hold very much more than this holds, and we enter numbers from little magnetic disks.

    JPE: In this memory device, in order to put numbers in it, you took a sheet of paper with the numbers on it, held it in front of you, transferred those numbers one at a time to these switches, the way I am imitating here. I'm just setting the numbers on one row here. To fill this table up completely, there are one hundred and four of these rows to be done, and in addition, there were three tables in some problems, so you had to do the thing I just did over three hundred times. As you can see, this took quite some time. There was a good chance for error. We made some of the rows red, every fifth one red, to facilitate knowing where you are, cutting down on errors, and we made every three columns a different color, to help know where you are in the columns. Today, of course, all this is replaced by a small chip which does the job faster and enormously cheaper, and we feed the information in from a magnetic disks that's worth a few cents. This is where the idea started.