The Eckert-Mauchly Computer Corporation
Building the Computer Industry
John Adam Presper Eckert Junior and John William Mauchly began work on designing their next computer before the construction of ENIAC had started, and they’d completed the design before ENIAC’s construction was completed. EDVAC was intended to be smaller, cheaper, faster, and more flexible than the machine before it, and the contract for EDVAC’s construction and delivery was signed in April of 1946. The initial budget was $100000 (about $1.5 million in 2024) but the total cost amounted to $467000 (about $7.3 million in 2024) which was similar to the ENIAC before it. Unlike ENIAC, the EDVAC would use what later became known as von Neumann architecture despite von Neumann not having designed it. The ENIAC used separate memory for data and instructions while the EDVAC used a single pool of memory for both that was built with one hundred twenty eight mercury acoustic delay lines with eight word capacity on each line yielding 5.6K RAM. Each word, in this case, was forty four bits. Additionally, EDVAC was fully binary. Stating that EDVAC was smaller than its predecessor was true, but this was still a very big computer. The EDVAC was comprised of five thousand nine hundred thirty seven vacuum tubes, twelve thousand diodes, and occupied nearly five hundred square feet. The EDVAC also improved on data input and output by making use of magnetic tape which was run at seventy five inches per second. While this machine was an improvement in some ways, it was a downgrade in others. First, the mercury delay lines had some thermal issues. One would initially imagine that vibration would be a problem, but it turned out that the design was good enough to prevent failures from vibration. Keeping the mercury warm was the major issue. This was eventually mitigated with a rather elaborate heating element not too dissimilar from an oven, and then the memory system required several thermostats and thermal buffers. Meanwhile, the processing system still required rather elaborate cooling. As for processing power, EDVAC could perform around three hundred forty multiplications per second, but addition, subtraction, multiplication, and division were all in hardware with both fixed and floating point capabilities. So, if one were using fixed point and doing just additions or subtractions, performance could be much higher at nearly twelve hundred operations per second. EDVAC consumed about fifty six kilowatts of power, and had a mean time to failure of approximately eight hours. The military’s report states that EDVAC averaged one hundred forty five operational hours in any given week, and this machine was attempted to be run twenty four hours a day with three eight hour shifts. The mercury delay lines were later replaced with magnetic drum memory. This increased memory latency but also increased total memory bandwidth.
Unfortunately, neither Eckert nor Mauchly would ever personally take part in the implementation of their design. The Moore School’s policies were changed around the same time and were they to remain at the University of Pennsylvania, they’d have been required to sign over the intellectual property rights for their inventions. Rather than do this, the two men resigned.
There are some sources indicating that the men sought to join John von Neumann at IBM, but this doesn’t seem entirely likely. From their own reports, Eckert and Mauchly do not have any nice words to say about von Neumann. Their early opinions of von Neumann appear to have been largely variations on indifference, and their later opinions of him were decidedly negative with Eckert having referred to him as a huckster.
Mauchly began researching the needs of potential computer users in the second half of 1944. Today, we know that he met with the US Census Bureau and the US Army Signal Corps, but he may also have had his first meetings with Prudential and AC Nielsen. With a sense of what needs potential computer users of the time would have, the two men chose to start their own company. The two were among a very short list of people with experience in what could scarcely have been called an industry at the time. They had possibly the most extensive knowledge of the market, and they were starting a new company. Exciting stuff! Except… no one else thought so. The two men were unable to get any funding at all. Electronic automatic computers were huge, expensive, costly to run, costly to staff, noisy, and exceedingly rare. Most of the organizations using computers were, at least in some way, connected to the military. To emphasize this prevailing sentiment of the era, Howard Hathaway Aiken (inventor of the Harvard Mark I) said in 1948 that there wasn’t really a substantial market for computers such that around a dozen would be sufficient for the scientific and engineering needs of the world. To say that Eckert and Mauchly were ahead of their time would be an understatement.
Eckert and Mauchly’s new computer company, Electronic Control Company, was founded on the 15th of March in 1946 at 1215 Walnut St. in Philadelphia. It was the first fully electronic computer company in the world, and the second private company to deal in automatic computers more generally (the first was Zuse Apparatebau established in 1941). Following the establishment of the company with Eckert as chief engineer and Mauchly as president and salesman, many of the bright minds who’d previously worked on ENIAC joined the company. The company’s first source of funds was Presper Eckert’s father, John Eckert, who loaned the new company $25000 (around $496000 in 2024). This really only worked because the majority of the young engineers who’d joined were willing to work for very little. This was, after all, a new and exciting technology, and the company was doing things no one else was. In September of 1946, the new company received a grant of $75000 (almost $1.2 million in 2024) from the National Bureau of Standards for a research project of Eckert’s that involved a computer using both mercury delay line memory and I/O via magentic tape.
On the 22nd of December in 1947, the company was incorporated as the Eckert-Mauchly Computer Corporation. Following incorporation, the company found their first customer in the Northrop Aircraft Company who purchased BINAC (BINary Automatic Computer). This was a rather weird machine. First, it was a bit-serial binary computer. This means that, like computers before it, it sent a single bit along a single wire at a time. Unlike its predecessors (except maybe EDVAC) it was a fully binary computer. Second, BINAC had 512 words of memory comprised of sixteen memory channels of thirty two words each where a word was thirty one bits. Third, for error checking, BINAC had a complete duplicate of every single component including the CPU and memory. All operations were compared and upon error could be rerun. As for program entry, it was tedious. All instructions entered into BINAC were done in octal machine code using an eight-key keypad. Fortunately, the computer could output data onto either a teletype or a wire recorder. The wire records could then be loaded as input. As a result, only the initial program entry need be done via the eight-key. BINAC had a CPU clock speed of 4.25 MHz and was capable of roughly 3500 operations per second.
While it is commonly claimed that BINAC was never made fully operational after delivery in September of 1949, this is accurate but not entirely truthful. The machine was successfully used for some problems, and in one case was able to work out deicing of an airplane in fifteen minutes while operators on electronic calculators had toiled away on the same problem for half a year. The primary issue wasn’t so much getting a single half of BINAC to run but rather in getting the two halves to work in concert. Roger Mills recalls:
The idea behind BINAC was to have all operations checked by running two sections of the computer independently and comparing each step on a high speed bus. Northrop engineers went back to Philadelphia for the acceptance tests. After the BINAC was shipped to Hawthorne, it sat out under a tarp for 6 months before it was assembled in its air-conditioned room. The poor engineers were constantly working to get BINAC to run. One side would be running while they worked on the other side. The two sides never worked together as long as I was there.
As for why the machine didn’t work as designed after delivery, there are three theories. The first, from EMCC, is that the machine wasn’t properly reassembled. The second, from Northrop, is that the machine wasn’t packed properly for shipping. The third, from me after reviewing statements about the machine from many sources, is that both of these are true and further that Northrop allowed the machine to sit for too long under a tarp in their parking lot likely damaging many components, and that they understaffed the group tasked with working on the machine. Northrop also wouldn’t allow any EMCC personnel into the Hawthorne site to assist their employees due to security concerns.
One innovation of BINAC that survives to the modern era is that of user manuals. Earlier machines were designed, implemented, and programmed by people who were intimately familiar with them and therefore didn’t exactly require user manuals. The BINAC was being delivered to a group who had been involved in neither its design nor its implementation but would be involved in its operation. Taking inspiration from the automobile industry, EMCC created a user manual for the BINAC and supplied this along with the machine.
At roughly the same time that BINAC was begun within EMCC, the company struck a deal with the Census Bureau to produce UNIVAC (UNIVersal Automatic Computer). The contract was signed in 1948 with the intent for the computer to be completed and delivered in time for the 1950 census. Four more UNIVAC contracts were secured shortly thereafter: the US Air Force, the US Army Map Service, Prudential, AC Nielsen.
While EMCC had quite a bit of technical talent, it had no business acumen. The BINAC was priced at $100000 but cost the company $278000 to build. UNIVAC’s first contract with the Census Bureau listed a price of $159000 (around $2 million in 2024), while the USAF and USAMS contracts listed a price of $250000 (about $3.2 million in 2024). These prices were well short of the costs of development and construction. The young company was quickly running out of money building BINAC and UNIVAC, and it was saved by Harry L. Straus who was the Vice President of the American Totalisator Company. He provided $500000 (about $6.4 million in 2024) in funding for which ATC received forty percent of EMCC, and Straus was named chairman. Unfortunately for EMCC, Straus perished in a plane crash in October of 1949 and ATC withdrew from EMCC. Once again, the company was struggling and in January of 1950 the company was put up for sale. Remington Rand purchased EMCC on the 15th of February in 1950 and EMCC became the Eckert-Mauchly Division of Remington Rand (EMDRR). Following purchase, Remington Rand provided the funding necessary to complete UNIVAC and subsequent UNIVAC contracts had much more realistic pricing of about $1.3 million (about $16.6 million in 2024). At this pricing, UNIVAC was far too expensive a machine for the majority of institutions who might require a computer. In an attempt to stop losing so much money, Remington Rand was able to convince Prudential and Nielsen to terminate their contracts.
When it was made available for purchase in 1951, UNIVAC was an improvement over earlier computers in almost every way, and it was better in some ways than many of the machines that followed (such as the 701, 702, and 650 from IBM). The first major innovation of UNIVAC that would influence all later computers came from Admiral Grace Murray Hopper. Hopper had joined EMCC in 1949 and stayed with the company after its purchase by Remington Rand. She felt strongly that computers ought to be easy to use, which led to the creation and use of the UNITYPER. This was a standard qwerty keyboard with ten-key that could enter data onto magnetic tapes. In addition to the UNITYPER, the supervisory control system also had a keyboard for input and corrections with the same type of keyboard along with all of the usual switches and blinken lights. This us brings us to the next major innovation, UNIVAC didn’t use punched cards. Input was done either via magnetic tape or via the UNITYPER, and output was done with a printer or magnetic tape (or both). Output could be made in real time and/or at the end of a batch job. If one had punched cards, there was an automated converter for card-to-tape conversions capable of processing roughly two hundred cards per minute (each card was actually read twice in that figure for error catching). Magnetic tapes could be rewritten, and therefore UNIVAC did have an optional tape-to-card converter if an organization wished to utilize cards for long term storage of applications or data. While it did have some card facilities, UNIVAC itself could not make any direct use of cards. These innovations greatly improved I/O speed as well as the programming tasks of the machine.
UNIVAC was a large machine with duplicated logic circuits for error checking. The mainframe was roughly fourteen feet long, eight feet wide, and eight and a half feet tall. The machine produced a prodigious amount of heat which required a chilled water and blower air conditioner for cooling. The CPU was built of fifty two hundred vacuum tubes and operated at 2.25MHz performing roughly 1905 operations per second. This was smaller than most other CPUs of the time, in both tube count and physical dimensions, due to the use of mercury delay line memory. The memory was organized into one hundred channels of ten words each yielding one thousand total words. The word size was seventy two bits with two instructions per word, or with twelve digits or characters per word. Together, in modern terms, UNIVAC had 9KB of RAM. There were more mercurial memories in the machine to act as I/O buffers, registers, and temperature controls. The complete machine with tape drives, control station, printers, and mainframe weighed in at roughly fifteen tons (without punched card converters), used eighteen thousand germanium diodes, consumed around one hundred twenty five kilowatts of power, and required raised false flooring to hide the connective cabling and plumbing.
Both mercury delay lines and vacuum tubes are temperature sensitive. As a result, the machine required a warm-up time of around thirty minutes before use. This long warm up was absolutely necessary if an organization wished to reduce failure rates for tubes and memories. Over time, burn-in procedures were established for tubes which increased the reliability of the UNIVAC. With tube burn-in practices combined with the lengthy warm-up, UNIVAC proved to be a very reliable and powerful computer with some installations seeing use for more than a decade.
With the first government contracts fulfilled, EMDRR wanted to get some advertising and increase sales. For this, they contacted CBS. UNIVAC serial number five was intended for the University of California Berkeley Radiation Laboratory (now the Lawrence Berkeley National Laboratory, but best known for its second location, Lawrence Livermore National Laboratory), but it had a job to do before it could go to its home. CBS used UNIVAC number five to predict the outcome of the 1952 US presidential election. The machine was still at the factory being tested, and the staff at EMDRR wrote a program that would predict overall election results given key precincts selected by CBS. On Tuesday, the 4th of November in 1952, the reporters at CBS News telephoned the data-entry clerks at EMDRR and provided the results as they came in. The clerks entered the results on the UNITYPER and ran the program. At 20:30 Eastern based upon 3.4 million votes, UNIVAC correctly predicted the election outcome with an Eisenhower victory. Remington Rand sold, built, and installed a total of forty six UNIVACs with the first being installed in 1952 for the US Census Bureau in Suitland, Maryland.
UNIVAC set several standards with the use of tapes, keyboards, printers, and raised floors, and the UNIVAC proved that computers could be far more useful and put to many more tasks than had previously been imagined. EMCC and UNIVAC essentially created the computer industry. Incidentally, the UNIVAC also created high level languages. It was on UNIVAC that Grace Hopper’s A compiler (1952) as well as the MATH-MATIC and FLOW-MATIC languages (1954) were initially written and run.
The IBM 650 or 701 are often cited as the first commercial and/or first mass-produced computers, but these honors go to BINAC and UNIVAC respectively. Eckert stayed with EMDRR and subsequent companies until his retirement. Mauchly left after his initial ten year contract expired. Eckert and Mauchly remained close friends. Admiral Hopper eventually left and joined Digital Equipment Corporation although her career is interspersed with commissions in the Navy. When she retired from the Navy (for the final time), she was the oldest active-duty commissioned officer in the US Navy at seventy nine, and her retirement ceremony took place on the oldest commissioned ship, the USS Constitution, aged one hundred eighty eight.