Unisys History Newsletter

Volume 5, Number 1  January 2001

UNIVAC I: The First Mass-Produced Computer

by George Gray

The computers of the late 1940s were all one-of-a-kind special projects done at research laboratories or companies closely tied to government funding. In 1946 J. Presper Eckert and John W. Mauchly started a company which had the goal of producing computers to be offered for sale in the commercial marketplace. They, too, were dependent on government funding to get started, but they wanted to reach customers in the larger business community, not just those involved in defense or other government work. After many years of struggle, they produced the UNIVAC I computer, which was delivered to its first customer (the U.S. Census Bureau) in 1951. Eventually, 46 UNIVAC I computers were built. Soon, other computers would be manufactured in much larger quantities, but from the perspective of the 1950s, this was mass production.

Establishing a Company

Eckert and Mauchly had been the principal designers of the ENIAC computer which was built at the University of Pennsylvania's Moore School of Engineering between 1943 and 1945 for the U.S. Army. The Moore School had also started working on an improved computer called the EDVAC. When World War II was over, many of the engineers at the Moore School left to take jobs elsewhere. Eckert and Mauchly resigned from the Moore School in March 1946 because of a dispute over patent rights, but stayed in Philadelphia to form a partnership named the Electronic Control Company. Their goal was to produce computers for both scientific and business use. This was a bold, indeed foolhardy, move, since hardly anyone besides Eckert and Mauchly believed that there would be a large demand for computers. The predominant opinion was that only the biggest government agencies and research institutes would ever have a need for a computer. One or two dozen machines would take care of all the scientific computation, while tabulating machines could continue to handle business processing.

The Electronic Control Company obtained a grant of $75,000 from the National Bureau of Standards in September 1946 for a research and study project involving a mercury delay line memory system and tape input/output devices. With the prospect of receiving some money, the company rented two floors of a building at 1215 Walnut Street in Philadelphia and began to hire employees. Eckert and Mauchly planned to expand the memory and tape efforts into the development of a complete computer system. It was originally referred to as an "EDVAC-type machine", but in May 1947 it was named UNIVAC for Universal Automatic Computer. The research project was followed in June 1948 by a design contract for $169,000. These amounts fell short of the actual costs, but Eckert and Mauchly hoped to recover the difference by sales of additional Univac’s. The company had a great deal of difficulty getting additional government grants or private investment. These years were the beginning of the Cold War between the U.S. and the Soviet Union, and problems arose with the security clearances of Mauchly and several other employees. The problems were eventually cleared up, but in the meantime the company lost potential sales opportunities with the Navy and a nuclear project at Oak Ridge, Tennessee.

The BINAC

As an interim measure to get some money, Eckert and Mauchly agreed in October 1947 to build a computer for Northrop Aircraft Company. It was called the BINAC (Binary Automatic Computer). They were supposed to complete it by May 1948, a hopelessly optimistic schedule. As the name indicated, BINAC performed binary arithmetic, rather than using decimal as had been done in the ENIAC. The word size was 31 bits: 30 bits for the value and one bit for the sign. The BINAC had two independent central processing units (CPUs), each with its own 512-word mercury delay line memory. The CPUs compared their results as a built-in double-checking mechanism. The BINAC used a primitive tape unit to input data and instructions. The processor had an accumulator register (A) and two other registers (L and R) which were used in multiplication, division, and some other operations. BINAC had a set of 16 instructions, which were called "orders," made up of a six-bit operation code and a nine-bit! field to hold the memory address.

The BINAC's first test program, 23 instructions in length, was run in March 1949. The BINAC and the EDSAC (Electronic Delay Storage Automatic Calculator), developed at Cambridge University, share the honor of being the first working computers to implement the stored-program concept. The BINAC was demonstrated for several months in Philadelphia and turned over to Northrop in August 1949, fifteen months behind schedule. It was taken apart and shipped to Northrop's headquarters in California, where it was reassembled and used for a short time. Some people have said that the BINAC never worked properly after its shipment to California, but two Northrop engineers later stated that it did.

Eckert-Mauchly Purchased by Remington Rand

The contract price for the BINAC was $100,000, but it cost the company $278,000 to build it. While they were very creative engineers, Eckert and Mauchly were terrible at estimating project costs. Isaac Auerbach, an employee who left the company in 1949 to work for Burroughs, was highly critical of their management abilities: "Neither Eckert or Mauchly in my opinion were competent managers, competent leaders, or competent executives, or understood business at all. They were visionaries, and they were brilliant technically, and they would not let somebody else run the side of the company in which they were inept." In December 1947, the company incorporated under the name Eckert-Mauchly Computer Corporation (EMCC).

After the completion of the BINAC, EMCC concentrated its efforts on the UNIVAC. The company contracted with the U.S. government to provide three computers, intended for the Census Bureau, the Air Force, and the Army Map Service, at a price of $159,000 for the first machine and $250,000 for the other two. Contracts for the UNIVAC were also signed with Prudential Insurance and the A.C. Nielsen Company. By the summer of 1948, EMCC had just about run out of money, but it was saved by the infusion of $500,000 from the American Totalisator Company, a Baltimore firm which made the machines (totalisators) for posting odds and reporting results at horse race tracks. Harry Straus, vice president of that company, felt that EMCC's work, besides being promising in general terms, might have some application in the race track business. Straus became chairman of the EMCC board and American Totalisator received 40 percent of the stock. Unfortunately, Straus was killed in an airplane crash in October 1949, after which American Totalisator provided no further support.

The contract prices for the Univac’s were so far short of the actual development costs that the company was again out of money. Eckert and Mauchly could not obtain any investment capital to keep the company going, so on February 1, 1950 they sold it to Remington Rand Corporation. A total of $100,000 was paid to EMCC stockholders, plus 49% of any profits over the next eight years that resulted from EMCC patents. Eckert remained with Remington Rand and its successor companies for many years, but Mauchly resigned in 1958 and went into business as a computer consultant. Neither man obtained great wealth from his leadership in developing the first working general-purpose electronic computers.

The First UNIVAC

Remington Rand provided the money to finish the UNIVAC. To reduce the financial losses, it cajoled Prudential and Nielsen into canceling their contracts. The first UNIVAC passed its formal acceptance test on March 29-30, 1951 and was turned over to the Census Bureau, which operated it in the factory for nearly a year. A formal dedication ceremony was held on June 14, but coverage in the general press was minimal. The following day, the New York Times ran a tiny, two-sentence article that referred to the UNIVAC as an "eight-foot-tall mathematical genius, designed to meet problems of the United States Census Bureau", but didn't even mention its name.

The central complex of the UNIVAC was about the size of a one-car garage: 14 feet by 8 feet by 8.5 feet high. It housed the mercury memory unit and all the central processing unit circuitry. The outside of the unit was composed of hinged gray metal doors that could be opened to give access to the circuitry racks. In the center of one of the long sides of the unit, there was a clear Plexiglas door to provide access to the center of the system: it was a walk-in computer. The vacuum tubes generated an enormous amount of heat, so a high capacity chilled water and blower air conditioning system was required to cool the unit. In addition to the central complex, there were eight UNISERVO tape drives, an operator console, and a console typewriter/printer. Originally printing was done offline by the UNIPRINTER, which resembled an overgrown typewriter with an attached tape drive. A much-needed 600 line per minute printer (at 130 characters per line) was added in 1954. The complete system had 5200 vacuum tubes, weighed 29,000 pounds, and consumed 125 kilowatts of electrical power.

The UNIVAC represented numbers in binary-coded decimal with six bits for each digit. It employed Excess-3 notation where the binary value was three greater than the actual number, so that zero was 000011, one was 000100, two was 000101, and so on. Excess-3 had been used in the Bell Telephone Laboratories Model I Relay Calculator built in 1940. Excess-3 was chosen for the UNIVAC because it simplified the complementing (making negative) of numbers and made the carries come out right for digit-by-digit decimal addition.

The Univac’s word size was 72 data bits, which held eleven digits plus a sign, plus one parity bit for each six data bits, giving a total of 84. The mercury delay line memory amounted to 1000 words. Besides numbers, the UNIVAC could represent alphanumeric data (letters of the alphabet and some punctuation marks) using six bits for each character with twelve characters to the word. Codes were assigned for the letters of the alphabet and punctuation marks, such as 010100 for A, 010101 for B, 010110 for C and so on.

The program instructions were six decimal digits (36 bits, excluding parity bits) long, so two instructions fit in each word. The first two digits of an instruction were the function code, the next digit was unused, and the last three gave the memory address. There were 45 different functions. Many of the function codes were mnemonic, that is, they tried to bear some relation to the operation to be performed. For example, A (still indicated by the bit pattern 010100) was the code for addition. Similarly, D was divide, S was subtract, and C meant to copy the contents of the A (accumulator) register into memory. All the functions didn't work out mnemonically: J meant to store the contents of the X register into memory. On the UNIVAC, addition could be done with just one register, in this case the A register, but other operations involved registers which were designated L and X. An add instruction, such as A 0503, would add the value at the stated memory location (503, in our example) to the value in the A register, leaving the result in the A register. The C instruction C 0504 meant to copy (store, in modern terminology) the value in the A register into memory location 504. On the UNIVAC, multiplication and division involved three registers. For example, the P multiplication instruction multiplied the value in the L register by the value in the stated memory location, giving a 22-digit product contained in the registers A and X. Tape input/output used two 60-word buffers designated I (input) and O (output). The input/output instructions provided for both forward and backward reading of tape. The read backward was particularly useful for sorting, where long strings of data were repeatedly written to tape and read back in through successive merges.

The computer had a high-degree of self-checking: all processing was done in duplicate by two sets of circuitry, and the results were compared to be sure they were identical. Donald Marquardt of DuPont recalled: "One of the big advantages of the UNIVAC was in fact the ability to rely on the accuracy of the numbers when they came out.... Now there were some other computers that I used during that same period where I would make two or three runs on the machine and come up with two or three [different] sets of numbers...."

The UNIVAC had the ability to store the control counter value in memory, making it possible for the flow of a program to go to a subprogram and then return to where it was in the main program. While the 72-bit word could accommodate numbers up to 11 digits, scientific calculations quite often involved larger numbers. To take an example from chemistry, Avogadro's number (the number of molecules in a mole of gas) is 6.02 x 10**23 ; computers could represent this in what is called floating-point format, where part of the word contains the value (6.02) and part of the word contains the exponent. Later computers would be designed with electronic circuits to perform calculations on numbers in floating-point format, but the UNIVAC did not have hardware instructions of this sort. Floating-point calculations could, however, be done by means of software subprogram, making it possible for the UNIVAC to do both scientific computation and business data processing.

Early in the design of the UNIVAC system, Eckert and Mauchly had recognized that for the computer to be useful in handling the large volumes of data used in many business applications, such as payroll of inventory control, it would need to have a high speed input/output system. Since punched cards would be slow, the company developed the UNISERVO tape units to be the primary input/output devices for the computer. Each unit was six feet high and three feet wide. The UNISERVO used metal tape: a 1/2-inch wide thin strip of nickel-plated bronze 1200 feet long. These metal tape reels were very heavy: not the sort of thing for an operator to drop on his or her foot! Data was recorded in eight channels on the tape (six for the data value, one parity channel for error checking, and one timing channel) at a density of 128 characters per linear inch of tape. The tape could be moved at 100 inches per second (as compared with 1.875 on today's cassette tape players), giving a nominal! transfer rate of 12,800 characters per second. Making allowance for the empty space between tape blocks, the actual transfer rate was around 7,200 characters per second.

No punched card devices were provided with the UNIVAC, so the UNITYPER data entry machine was developed. The data entry clerk typed on a keyboard, and the UNITYPER recorded the values on a reel of metal tape. This lack of integration with punched card systems became a marketing handicap. Many prospective customers already had significant investment in tabulating card systems. When IBM entered the computer business, it made sure that it offered computers that fit easily into existing card processing installations. To fill this gap, Eckert-Mauchly developed a stand-alone card-to-tape unit, which could process 100 cards per minute. Since Eckert-Mauchly was an independent company at the time the design of the card-to-tape converter was done, it naturally followed the market and built a machine that handled IBM's 80-column cards. Sometime after the acquisition by Remington Rand, a version to handle 90-column cards was developed.

Remington Rand Expansion and Merger

In December 1951 Remington Rand acquired another pioneering computer company, Engineering Research Associates (ERA) in St. Paul, Minnesota. In addition, the company had its own tabulating machine research and development effort in Norwalk, Connecticut. These three organizations (Eckert-Mauchly, ERA, and Norwalk) continued to function with a high degree of autonomy, indeed rivalry, for several years. At some point in 1953 or 1954, the company decided to use the UNIVAC name for the products of all three groups, so the original UNIVAC computer was re-designated the UNIVAC I and ERA's 1101 computer became the UNIVAC 1101. In 1955, Remington Rand merged with Sperry Gyroscope to form Sperry Rand Corporation.

Later UNIVAC I Installations

The Census Bureau operated the first UNIVAC I in the Philadelphia factory for nearly a year before it was dismantled, shipped to the Census office in Washington, and reassembled. Remington Rand built the next five UNIVAC Is for other U.S. government agencies: the Air Force, the Army, two for the Atomic Energy Commission (AEC), and the Navy. The initial pace of installations was slow. Prior to the buy-out, EMCC had been planning to set up production capacity sufficient to build one UNIVAC I per month, but Remington Rand cut the plan back to one machine every two months. In fact, the second machine (Serial No. 2), for the Air Force Comptroller was not completed until February 1952, eleven months after the first. After some time in the factory, it was shipped to the Pentagon, where it was in full operation by June. Its first task was to run a linear programming model to do logistics calculations for war planning. (When it outlived its usefulness there in 1958, the machine was! shipped yet another time to the Air University at Maxwell Air Force Base in Alabama.) Serial No. 3 was operated in the factory by its user agency, the Army Map Service, from its acceptance in April 1952 until September 1952, when it was moved to Washington. Full operation at the Army Map Service got underway in January 1953; in the interim the Army used Serial No. 4 in the factory. The Atomic Energy Commission received the next two machines, installing Serial No. 4 at New York University and Serial No. 5 at the University of California Radiation Laboratory. Serial No. 6, completed in April 1953, was built for the Applied Mathematics Laboratory of the Navy's David Taylor Model Basin. At its formal dedication on October 16, 1953, the UNIVAC ran a program that played "Anchors Aweigh" on the speaker. Another program pitted the UNIVAC against human players in a mathematical game called NIM; naturally, the UNIVAC beat all the humans, including John Mauchly.

The CBS television and radio networks used Serial No. 5 to predict the outcome of the 1952 presidential election. The computer had just been completed and was still being tested at the Philadelphia factory. Remington Rand staff wrote a program that would make a prediction based on results from key precincts selected by CBS. Like many programming projects ever since, it turned out to be a bigger task than expected, and the programming staff had to be increased from two to six people in order to get finished on time. On election day, CBS News reporters phoned the data to Philadelphia where data-entry clerks entered it in triplicate into the UNIVAC via three UNITYPER key-to-tape machines. The program compared all three versions in order to catch typing errors. At 8:30 P.M. Eastern time on the basis of 3.4 million votes, the program predicted an Eisenhower victory by a wide margin, but CBS did not announce it for over an hour, waiting for more returns to be sure that the UNI! VAC was right. Jack Gould, the television writer for The New York Times, was not impressed with either the UNIVAC or the much smaller Monrobot computer used by the NBC network: "Both gadgets were more of a nuisance than a help." However, the CBS television announcers praised the UNIVAC highly, pointing out that while it had produced an accurate prediction, they had been afraid to believe it.

This was enormously favorable publicity for the UNIVAC I, and for several years the name UNIVAC was synonymous with computer in the public mind. An instance of this was noted as late as the 1974 football season, when Datamation reported that the television announcer Alex Karras, amazed at his partner's vast knowledge of football statistics, said to Howard Cosell: "You're a real UNIVAC, Howard!" Datamation did comment that Karras was a little behind the times. In a series of short stories written during the 1950s, the famous science fiction writer Isaac Asimov described a computer called Multivac, whose name was an obvious variation of UNIVAC. This public confusion of UNIVAC with computer was extremely galling to the leadership of IBM.

In 1953, General Electric became the first non-government entity to order a UNIVAC I, receiving the eighth one built. The seventh had been installed at the Remington Rand sales office in New York City. In those days when a suburban house could be bought for $5000, purchase of a $1 million computer was not a decision to take lightly. General Electric hired the consulting firm Arthur Andersen to do a feasibility study and analysis. The study took from April to July of 1953 and recommended the UNIVAC I over a prototype of the IBM 702 computer. In the area of punched card tabulating equipment, IBM had the upper hand over Remington Rand, and many people looked on Remington Rand with scorn. John Swearingen of GE remembered that when his boss Roddy Osborn accepted the Arthur Andersen recommendation there was uproar. People asked: "How could you go with that outfit? How did you not pick IBM?" IBM pressured GE's top management to reverse the decision, but management supported! Osborn, and Remington Rand began installing the UNIVAC at GE's appliance factory in Louisville in January 1954.

The installation team ran into many problems, since some key components necessary for GE's commercial environment, such as an improved line printer and the card-to-tape converter, were not ready on schedule. Willis Drake, a veteran of the ERA division, was dispatched to Louisville to expedite the installation, and he spent several months getting the machine to the point where it would work. The first application was the factory payroll, an extremely complex task, since the workers were paid a combination of hourly wages plus piecework compensation plus various incentives triggered by exceeding production quotas. Up to this time, most of GE's payroll had been done by hand; the home laundry division was the only one to have moved to a tabulating card system. Prior to the installation the programmers thought that the 10,000-person payroll could be run in four hours; in fact, the first payroll run took 44 hours. As the programmers gained experience and refined the program, t! he time was eventually cut to 20 hours. Later on, GE added a materials inventory control system, which was successful in reducing the factory's inventory costs by $1 million.  

The UNIVAC I soon found a home in insurance companies. Pacific Mutual Insurance first expressed interest in September 1952, and the following year sent two employees to UNIVAC programming school. Pacific Mutual evaluated both the UNIVAC I and IBM 702. The UNIVAC had the advantage that it was actually installed and working, while no 702s had been shipped. Pacific Mutual considered the UNIVAC to be technically superior and finally placed its order in 1954. The computer was actually installed in August 1955. Delays like this were usual, since the Philadelphia factory had limited capacity. However, customers could make a virtue of necessity and use the interval for the lengthy process of preparing programs. Since computers were new and no one knew what to expect, Pacific Mutual went to great lengths to inform its employees of the progress of the computer project. The company also provided retraining for employees whose jobs were eliminated by the introduction of computerized processes. In the meantime, Univac’s had already been installed at Metropolitan Life in New York City and Franklin Life in Springfield, Illinois. Franklin Life had high expectations for its machine. The company president told the New York Times: "The advantages of the new installation to the Franklin Life are obvious and numerous. Statistical data for the guidance of the management will be forthcoming much more rapidly than heretofore, making possible almost immediate decisions or policies which otherwise would necessarily have to be delayed for weeks or months." Franklin Life must have been pleased, because it later added a second UNIVAC I.

DuPont bought the twelfth UNIVAC, and it was delivered in September 1954. At DuPont, unlike the insurance companies, the focus was on scientific programming rather than business processing. The company's accounting division was a stronghold of IBM punched card tabulating equipment. However, chief engineer Granville Reed decided that the UNIVAC would be a good for solving computational problems which were too lengthy to be done by hand in a reasonable amount of time. Donald Marquardt of DuPont recalled a mathematician in the Army who took several weeks to do one ten-variable multiple regression by hand with a desk calculator. Once a program was written, such problems could be solved in a matter of hours or even minutes. Research groups at DuPont used the UNIVAC for statistics, economics, and project engineering. One project involved finding numerical solutions to differential equations that predicted the breaking strengths of fibers under various loads.

The later UNIVAC Is sold for around $1,250,000 price to $1,500,000. The customers included other insurance companies (John Hancock in 1955, two additional machines at Metropolitan Life in 1956, and one to Life and Casualty of Tennessee in 1956), various manufacturing corporations, (U.S. Steel, Westinghouse Electric, Sylvania, and Carborundum), and more government agencies (Navy Bureau of Ships, Air Materiel Command, and Internal Revenue Service). The Census Bureau got a second machine in October 1954. One significant weak spot was the educational market. The UNIVAC I was too expensive for most universities to buy or rent, and Sperry Rand's financial condition was not so strong that it could give many Univac’s away. It did donate Univac’s to Harvard University (in 1956) and to the University of Pennsylvania and Case Institute of Technology in Cleveland (in 1957).

One of the most interesting uses of the UNIVAC I was to prepare the first concordance of the Revised Standard Version of the Bible in 1956. John W. Ellison, rector of an Episcopal church in Massachusetts, had been given the task, and he decided to use computer facilities donated by Franklin Life Insurance and the Remington Rand service bureaus in New York and Los Angeles. The entire text of the Bible was keyed in over a period of six months by a staff of five women using UNITYPERs, while another group of ten entered it on punched cards, which were then transferred to tape by the card-to-tape converter. Each version of the text occupied five reels. The two versions were compared to eliminate errors. The main keyword extraction run took 250 hours and produced 26 reels, and the final sort into alphabetical order required another 50 hours. A similar concordance published in 1894 had taken 30 years to do by hand.

In all, 46 UNIVAC’s were produced. Just about every customer was very pleased with the UNIVAC I. Commonwealth Edison in New York was the only company known to have rejected a UNIVAC I after it was installed, and that is attributed to internal company politics, rather than any deficiency of the computer. The UNIVAC I was also the machine on which much of Grace Hopper's pioneering higher-level language development was done. Some of the UNIVAC Is remained in service for a long time. The Census Bureau retired its two machines in 1963. Sperry Rand itself used two of them in Buffalo until 1968, and Life and Casualty of Tennessee kept its UNIVAC I until 1970, for over thirteen years of service.

Trademarks

Unisys, UNIVAC, and UNISERVO are registered trademarks of Unisys Corporation.

Copyright 2001 by George Gray