ENIAC and the First Electronic Computers

Zusammenfassung

On February 14, 1946, the Moore School of Electrical Engineering at the University of Pennsylvania unveiled a machine that filled an entire room, weighed thirty tons, and consumed as much electricity as a small neighborhood. The ENIAC — Electronic Numerical Integrator and Computer — was the world’s first general-purpose electronic computer, capable of performing five thousand additions per second at a time when a skilled human with a mechanical calculator could manage a few dozen per minute. It was the product of a wartime contract, two visionary engineers, and six women mathematicians whose programming work would be forgotten for fifty years. It was also the machine that ignited a patent dispute whose resolution would, three decades later, strip the entire American computing industry of its founding intellectual property.

The Problem That Built the Machine

The immediate origin of ENIAC was not scientific curiosity but artillery.

By 1942, the Ballistic Research Laboratory (BRL) at the Aberdeen Proving Ground in Maryland faced a problem of overwhelming scale. Every new artillery piece — every new shell type, every new elevation range, every new atmospheric condition — required a ballistic firing table: a precalculated grid of trajectories telling gunners where to aim under each possible combination of variables. A single firing table required roughly 3,000 trajectory calculations. Each trajectory took a trained human mathematician — they were called “computers,” a job title — about twenty hours to compute by hand or with an electromechanical desk calculator. The BRL employed hundreds of such human computers and was still falling behind.

Herman Goldstine, an Army lieutenant with a mathematics PhD, was the BRL’s liaison to the University of Pennsylvania. In the spring of 1943, he encountered John Mauchly, a physicist who had been arguing for years that vacuum tubes could do arithmetic electronically — at speeds no mechanical device could approach. Mauchly and his collaborator J. Presper Eckert, a twenty-four-year-old graduate student with exceptional engineering instincts, had been sketching a design. Goldstine recognized immediately what it could do for the BRL’s backlog.

On June 5, 1943, the Army signed a contract with the Moore School for $61,700 (later expanded to $486,804) to build what the contract called an Electronic Numerical Integrator and Computer. Mauchly would provide the conceptual framework and application theory; Eckert would build it. They had two years.

The Machine: Anatomy of a Giant

The machine that emerged between 1943 and 1945 was unlike anything previously constructed.

ENIAC occupied 1,800 square feet of floor space — roughly the footprint of a large two-bedroom apartment. It stood eight feet tall and extended nearly a hundred feet in its horseshoe configuration. It weighed 30 tons. Its components included:

17,468 vacuum tubes (the fundamental switching elements, prone to failure)
7,200 crystal diodes
1,500 relays
70,000 resistors and 10,000 capacitors
Approximately 5 million soldered joints

  ENIAC — Physical Layout (simplified):

  ┌──────────────────────────────────────────────────────────┐
  │  Cycling unit │ Master programmer │ Printer               │
  ├───────────────┴────────────────────┴──────────────────────┤
  │                                                            │
  │  [Accumulator 1] [Accumulator 2] ... [Accumulator 20]     │
  │  (each: a 10-digit decimal register with its own adder)   │
  │                                                            │
  │  [Multiplier]  [Divider/Sq.Root]  [Function Tables ×3]    │
  │  [Constant Transmitter]  [Input/Output units]             │
  │                                                            │
  └────────────────────────────────────────────────────────────┘
  Panels connected by ~500,000 hand-soldered wire connections
  Programs stored as physical cable configurations + switch settings

ENIAC operated in decimal (not binary), using ten-state ring counters to represent digits — an engineering choice driven by the BRL’s requirement for direct compatibility with existing ballistic tables. It had 20 accumulators, each capable of storing and operating on a ten-digit number. A multiplier unit handled multiplication by repeated addition. Three function table units allowed read-only lookup tables to be hand-set on rotary switches — a crude but effective substitute for stored data.

Its operating speed was startling: 5,000 additions per second, 357 multiplications per second, 38 divisions per second. A ballistic trajectory that took twenty human-hours now took thirty seconds. The machine was, by a factor of roughly 1,000, the fastest calculator on earth.

Power consumption was substantial — roughly 150 kilowatts — and the heat generated by 17,468 vacuum tubes required industrial-scale cooling. Tube failures were the operational nightmare of the early years. At those tube counts, statistical failure rates meant one tube burned out roughly every two days. Eckert’s engineering discipline kept the failure rate lower than expected through careful de-rating — running tubes at well below their rated voltages — but the machines still spent significant time in maintenance.

ENIAC ran its first successful test calculation on November 10, 1945 — three months after Japan’s surrender. It was formally unveiled to the public on February 14, 1946.

The Six Programmers Nobody Photographed

Among the most consequential — and most obscured — facts about ENIAC is who programmed it.

Programming ENIAC was not a matter of writing code. There was no code, no stored program, no software in any modern sense. Programming meant physically configuring the machine: routing cables between its function units, setting thousands of switches on its panels, and establishing the sequence of data flows that would carry a computation from input to output. To program a new problem, operators first had to understand the problem mathematically, decompose it into operations ENIAC could perform, and then translate that decomposition into a physical cable-and-switch configuration. Getting it right was equivalent to building a custom circuit for each application.

The Army assembled a team of six women mathematicians for this task:

Kay McNulty (later Mauchly Antonelli) — hired from a team of human computers at the Moore School
Frances Bilas (later Spence) — a mathematics graduate from Chestnut Hill College
Jean Jennings (later Bartik) — who had graduated first in her class in mathematics from Northwest Missouri State
Betty Snyder (later Holberton) — who had been told by an adviser that women should be teaching rather than doing mathematics, and ignored him
Marlyn Wescoff (later Meltzer) — hired initially as a human computer at the BRL
Ruth Lichterman (later Teitelbaum) — a mathematics graduate from Hunter College

They had no programming manual. There was no programming manual. They learned ENIAC from its electrical blueprints, tracking current through circuits to understand what each panel did. They were not permitted inside the machine room during the classified period of its development. When they were finally given access, they discovered that several circuits were wired incorrectly — and fixed them.

Forgotten at the Press Conference

At the February 1946 public unveiling, the six women were present. They were not introduced. The press photographs show men in suits standing beside the machine’s panels; the women who had configured those panels for the demonstration were not identified in contemporary coverage. When a celebratory dinner was held for the ENIAC team, the women were initially barred from attending — they were eventually admitted after protest, only to be seated separately. Their names did not appear in popular histories of computing for decades. Betty Holberton went on to co-design the COBOL and FORTRAN programming standards. Jean Jennings Bartik led the team that converted ENIAC to a stored-program machine in 1948. They received formal recognition — the Women in Technology International Hall of Fame — only in 1997.

For the public demonstration, the team programmed ENIAC to compute ballistic trajectories in real time — faster, it was noted for the press, than the shell itself was traveling. The calculation that had defined the project took thirty seconds. The press called ENIAC a “Giant Brain.” The team that had told the brain what to think was not mentioned.

Mauchly, Atanasoff, and the Question of Invention

The origin of ENIAC’s key electronic concepts was contested from the beginning.

John Vincent Atanasoff was a physics professor at Iowa State College who, from 1937 to 1942, built with his graduate student Clifford Berry a device they called the Atanasoff-Berry Computer (ABC). The ABC was not a general-purpose machine — it was designed specifically to solve systems of simultaneous linear equations — but it incorporated several ideas that were genuinely novel: it operated electronically (vacuum tubes), computed in binary, used capacitor-based regenerative memory to store intermediate results, and separated computation from memory.

The ABC was never completed. Iowa State did not patent it. When the United States entered the war, Atanasoff was transferred to other work. The machine was partially dismantled and eventually destroyed when the university repurposed the room.

In June 1941, John Mauchly visited Atanasoff at Iowa State and spent several days examining the ABC and discussing its principles. Mauchly later said he found the ABC unimpressive and that his own ideas about electronic computation predated the visit. Atanasoff said Mauchly had extensively studied the machine and the documentation Atanasoff had prepared.

Eckert and Mauchly’s patent application for ENIAC was filed in 1947. It was assigned to the Sperry Rand Corporation after Remington Rand acquired their company. By the late 1960s, Sperry Rand was demanding licensing fees from every computer manufacturer in the United States on the basis of the ENIAC patent — effectively taxing the entire industry.

Honeywell, rather than pay, sued. The case — Honeywell Inc. v. Sperry Rand Corp. — was decided in 1973 by U.S. District Judge Earl Larson after 135 days of trial and examination of thousands of documents and depositions. Judge Larson found that Mauchly had derived fundamental concepts from Atanasoff’s work and that the ENIAC patent was invalid.

The Patent That Died

The ruling in Honeywell v. Sperry Rand (1973) did not resolve who invented the electronic computer. It resolved a legal question: whether the ENIAC patent was valid. Judge Larson found it was not — partly because Mauchly’s 1941 visit to Atanasoff constituted prior art disclosure, and partly on other technical grounds including prior public use. The practical effect was that electronic computing became unpatented: no single entity held intellectual property over the basic concepts. This was probably good for the industry. The engineering community, meanwhile, continues to debate Atanasoff’s place in the hierarchy of invention. The ABC was not general-purpose, not programmable, and not completed. ENIAC was. The ideas Atanasoff developed — binary arithmetic, electronic switching, regenerative memory — were real contributions, but the leap from the ABC to ENIAC was not incremental.

The Transition: From Wiring to Stored Programs

By 1947, ENIAC was operational at Aberdeen Proving Ground, computing firing tables as originally intended. But the machine’s future was already being redefined.

John von Neumann had entered the picture in 1944 through Herman Goldstine, who had encountered von Neumann on a train platform and recognized him. Von Neumann was drawn into the ENIAC project as a consultant, quickly grasping both its capabilities and its limitations. The central limitation was the one that had defined it from the start: the program was in the wires, not in the memory. Changing programs took days of physical reconfiguration.

Von Neumann’s insight — that programs could be stored as data in the same memory that held numbers — was not wholly his own. Eckert and Mauchly had been thinking along similar lines. But it was von Neumann who wrote it up formally, in the June 1945 “First Draft of a Report on the EDVAC” — a document that circulated widely under his name alone and became the founding specification of stored-program computing. The credit dispute this created is examined in detail in The Von Neumann Architecture.

What is less often noted is that ENIAC itself was eventually converted. In 1948, Jean Jennings Bartik — one of the original six programmers — led the team that modified ENIAC to operate as a stored-program machine. The conversion involved permanently rewiring a portion of the function tables to serve as a primitive read-only memory for instructions. ENIAC ran its first stored-program calculation in September 1948 — a Monte Carlo simulation of nuclear chain reactions for the Los Alamos laboratory. It was not a clean solution (the “program memory” held only a few dozen instructions), but it demonstrated the principle and extended ENIAC’s operational life.

UNIVAC: The Commercial Leap

While ENIAC continued operating at Aberdeen, Eckert and Mauchly had already moved on.

In 1946, they left the Moore School — in part over a dispute with the university about who owned patents on inventions developed under government contract — and founded the Eckert-Mauchly Computer Corporation (EMCC) in Philadelphia. Their goal was the UNIVAC: Universal Automatic Computer, a stored-program machine designed from the start for commercial data processing rather than ballistic calculation.

UNIVAC incorporated all the architectural lessons of ENIAC and EDVAC:

Stored-program operation (programs and data in the same mercury delay-line memory)
Binary internal representation with decimal I/O
Magnetic tape for mass storage, replacing the punch cards of ENIAC
A design explicitly intended for business and statistical applications, not just scientific computation

Funding was chronic problem. EMCC sold promised delivery contracts before building capacity; the Census Bureau committed to a machine in 1946 for delivery in 1949. The machine was late. Eckert and Mauchly, brilliant engineers but not businessmen, ran out of money. In 1950, Remington Rand acquired EMCC.

The first UNIVAC I was delivered to the U.S. Census Bureau on March 31, 1951 — the first commercial electronic computer delivered to a non-military customer in the United States. Forty-six UNIVACs were ultimately built and sold, at prices between $1 million and $1.5 million each. In November 1952, a UNIVAC connected to a CBS television broadcast correctly predicted Dwight Eisenhower’s landslide presidential victory from early returns — a prediction the network initially suppressed because it seemed implausibly large. The broadcast made UNIVAC a household word.

Dead End: The Wire-Program Machine

ENIAC itself was a magnificent dead end.

Its decimal arithmetic, its physically wired programs, its room-scale engineering — all of it was superseded within a decade. No subsequent computer adopted ENIAC’s architecture. The decimal ring counter approach to arithmetic was abandoned for binary representations that mapped cleanly to transistor states. The function-table programming model was abandoned for stored programs that could be loaded in seconds and modified without touching hardware. The vacuum tube was abandoned for the transistor.

What ENIAC demonstrated was not an architecture but a possibility: that electronic switching devices could compute, reliably and fast, and that the resulting machines were practically useful. This was not obvious in 1943. Skeptics argued that 17,000 vacuum tubes would fail so frequently that the machine would spend more time in repair than in operation. Eckert’s de-rating strategy proved them wrong. ENIAC ran for nine years — from 1945 to 1955 — a productive operational life for a first-generation machine.

The Erasure of the Women

ENIAC’s most significant dead end was not technical but historical: the near-total erasure of the six women who programmed it. For decades, popular histories described the women in photographs as “refrigerator ladies” — models hired to pose beside the machine for promotional photographs. They were not models. They were the people who made the machine compute. Betty Holberton went on to design the Sort-Merge Generator, one of the first software tools, and helped draft the COBOL and FORTRAN standards. Kay McNulty became John Mauchly’s wife and continued working on computing for decades. Jean Jennings Bartik led the stored-program conversion. Their systematic exclusion from computing history was itself a kind of technological failure: it deprived subsequent generations of role models and misrepresented what the field required and who could do it.

ENIAC was decommissioned on October 2, 1955, after running its last calculation at the Aberdeen Proving Ground. Parts of it are preserved at the Smithsonian Institution and the University of Pennsylvania. It ran for 80,223 hours of operation. In that time it solved ballistic trajectories, modeled nuclear weapons, ran weather simulations, and answered the essential question that justified its construction: yes, it is possible to build a machine that computes faster than any human can.

Everything that followed was built on that answer.