ENIACU. S. Army

Dr. John von Neumann speaking on the occasion of the first public showing of the IBM Naval Ordnance Research Calculator (NORD) December 2, 1954. The complete speech appears in von Neumann's collected works, vol. V, MacMillan 1961.

I digitized it from a cassette tape provided by Dr. Goldstine. I made a DAT copy of the cassette at 48,000 kHz stereo 16-bit linear encoding, Which I then loaded into an SGI workstation and used SOUNDFILER to crunch it down to 8 kHz mono in 8-bit mu-law encoding.

- Part 1 2.1M
Mr. President, Mr Chairman, President ?Church, Mr. and Mrs. Watson, ??? Julia, ladies and gentlemen, let me thank you first of all for a very kind introduction. I don't know how to comment on it.

I first of all hope that I'm, and I know for sure that I'm expressing the same sentiment which everybody who has been involved in this thinking about computing machines, using computing machines, expecting to use computing machines, probably feels at this moment, namely, a sense of an obligation to express our admiration and our felicitation both to the U. S. Navy and to the IBM Corporation for having asked for the NORD, for having produced it, for having in common deliberations and decisions had to set standards as remarkably high as they were set ... and for having brought off this really remarkable achievement.

I will say a few things about the importance of NORD, about what we hope it will do, and what it's significance in this development is. To talk about the significance of this development one first has to visualize what it's position in this development is.

We have reached a stage where we can use it's basic calibration as the achievement of by-gone days

[Lots more]

the one average figure: it's multiplication. ...how many of the subordinate operations one counts in but it's something by and large of the order of one twenty-thousandth of a second. This includes already something more than the multiplication itself.

- Part 2 1.7M
One might ask why this speed is needed, and what the problems are that Most of us have asked this question for years, most of us have answered it by now to our own satisfaction, so I will not repeat the familiar arguments. In speaking of problems which can be solved with NORD. Let me perhaps mention one or two with which I am particularly familiar. NORD being a machine of very high speed and a very large memory, is clearly suited for problems where large amount of data needs to be processed. those in geophysics. It is quite instructive what a machine of the type of NORD. The air, water, and earth. Speaking of air in the first place, the phenomena of the atmosphere. it is worthwhile to estimate what NORD could do by We know today that we can predict by calculation the weather over an area like that of the United States for a duration like 24 hours in a manner which from the hydrodynamicist's point of view is limited, because you only consider one level in the atmosphere, the mean motion of the atmosphere, an average. We know that these predictive results are by and large about as good as what an experienced subjective forecaster can do -- which is quite good. This kind of a calculation, from start to finish, with NORD would take about half a minute.

- Part 3 2.2M We know that this calculation can be ?? You can not refine this mathematically ad.infinitum, because once the mathematical precision has reached a certain level, it begins to loose it's significance because the physical assumptions which enter are no longer as good. In atmospherics, in the simple descriptions of the atmosphere, this level as we now know is reached when you deal with approximately three or four levels in the atmosphere. This is a calculation which NORD would probably do for 24 hours ahead in something of the order of 2 to 4 minutes. We know a calculation covering the entire globe which are to make meteriological forcasts for longer periods -- for periods during which the main circulational pattern that one in interested in are set up, which are periods like 30 to 60 days, that such calculations are probably possible, but that surely we have to consider areas much larger than the United States because in the duration for like 30 days, in fact for much shorter durations than 30 days, influences from more remote parts of the globe interact. We also know that interaction between the northern and southern hemisphere is not very strong, therefore one can probably limit the calcuation domain to one hemisphere. Such calcuations have so far only been done in very tentative and primitive ways, and all those who have worked on these problems have really worked on them so far in the spirit of getting oriented. One serious reason for people going easy on these problems is that even with the best available computing machines they are still a very large problem, [1:58] and when you deal with a new problem, you must resign yourself to solving it first a few dozen times the long way before you gradually find out by trial and error and by coming to grief what a reasonably good way is. Consequently one will simply not do it unless one can make the individual tests practically. [2:20] A calculation of this order on NORD goes, I would say something of the order of magnitude of 24 hours computing time. This can be off by a factor of two or three one way or the other, but that is the order of magnitude, which means that it is accessible for research purposes. I mention another area, namely calculations relevant to the ocean. complete calculation of the tidal motion in the entire oceanic system I will not try to put numbers in hours and days on calculations of this type because one has to go into considerable details for evaluation However, there is no question that with a machine like NORD this is maybe for the first time where the calculation is and therefore with all the trials and errors practical and feasible.
- Part 4 2.3M
To mention still another area, namely, things related to the earth, it has been realized for some time that the hydrodynamics of the liquid core of the earth It seems extremely likely that this is responsible for the phenomena of Another category of problems with which the so-called experts have been familiar are statistical experiments on complicated the power of the computing machine makes it possible to compute the outcome of something that you might do before because you can exclude 50%, 80%, 90% of what won't work on the basis I would like to add that this can be pushed further. There are complicated processes which are not exactly mechanical combat operations, or more simply logistical operations. Calculate what is going to happen because the event also depends on what somebody else will do, for instance or some other factor involving human events or vulnerabilities which are statistical go through a calculation a hundred times give you the correct statistical pattern and then get oriented how the decisions should be made. This has been done on minor scales, medium scales, before. nobody in planning a program, a research program, likes to commit himself to a particular plan which If this involves tieing up everyone for half a year, If one can do these things rapidly, one will be bordering on this kind of "more quickly how to do this", so even finding how to do these things requires very fast equipment, in addition to which, it can only be done in a routine way, and in the absolute manner in which it ought to be done, if such equipment is available.

Again here the importance of machine in the class of NORD, and specifically of NORD, will be quite, will be important (?).

- Part 5 2.3M, [6:56]
The same thing goes for various other calculations involving
catchwords of linear programming, non-linear programming
In speaking of the importance of NORD I would also like to add
and I tried to give some of the reasons for
Those of you present who are
who have lived and suffered with computing machines
you know what kind of regime it is to
I'm sure that you appreciate that
This machine has been assembled less than two months ago,
two weeks ago
yesterday ran for four hours
it's completely fantastic on an object of this size,
I doubt it has ever been achieved before, and it is
an enormous reassurance on the state of the art and
regarding the complexities which one will be able to go in the future,
that this has been achieved.
This is a machine of about 9 thousand tubes and 25 thousand diodes. These numbers are very high, but such numbers have occured before. But machines of this type in the past took several years to break in.

- Part 6 635K [33 seconds]
The last thing, which is very important, is said in fewer words, but I think that it is none the less important. And it is this: In planning new computing machines, in fact, in planning anything new, in trying to enlarge the domain of parameters with which one can work, it is of course customary and very proper that one should consider what the demand is, what the price is, whether it will be more profitable to do it in a bold way than in a cautious way, and so on. This type of consideration is necessary -- the world would very quickly go to pieces if these rules were not observed in 99 cases out of a hundred.

It is terribly important that there should however be one piece in a hundred where it is done differently.

And that one uses the definition which Dr. Haven (?) pointed out 20 minutes ago, namely to occasionally do what the U. S. Navy did in this case and what IBM accepted in this case: to write a specification essentially to build the most powerful machine that is possible in this day with the present state of the art. I just hope that this will be repeated very soon and will never be forgotten.

Up |