History of computing hardware

Ancient and medieval

Devices have been used to aid computation for thousands of years, often using one-to-one correspondence with fingers. The earliest counting device was probably a form of tally stick. The Lebombo bone from the mountains between Eswatini and South Africa may be the oldest known mathematical artifact.^[2] It dates from 35,000 BCE and consists of 29 distinct notches that were deliberately cut into a baboon's fibula.^[3][4] Later record keeping aids throughout the Fertile Crescent included calculi (clay spheres, cones, etc.) which represented counts of items, probably livestock or grains, sealed in hollow unbaked clay containers.^[b][6][c] The use of counting rods is one example. The abacus was used early for arithmetic tasks. What is now called the Roman abacus was used in Babylonia as early as c. 2700–2300 BC. Since then, many other forms of reckoning boards or tables have been invented. In a medieval European counting house, a checkered cloth would be placed on a table, and markers moved around on it according to certain rules, as an aid to calculating sums of money.

Several analog computers were constructed in ancient and medieval times to perform astronomical calculations. These included the astrolabe and Antikythera mechanism from the Hellenistic world (c. 150–100 BC).^[8]

A Greek bronze combination lock from the Augustan or Hadrianic period operated on a primitive form of mechanical logic: the central bolt was physically blocked from retracting until the notches of two independent rotary dials were correctly aligned.^[9] In Roman Egypt, Hero of Alexandria (c. 10–70 AD) made mechanical devices including automata and a programmable cart.^[10] The steam-powered automatic flute described by the Book of Ingenious Devices (850) by the Persian-Baghdadi Banū Mūsā brothers may have been the first programmable device.^[11]

Other early mechanical devices used to perform one or another type of calculations include the planisphere and other mechanical computing devices invented by Al-Biruni (c. AD 1000); the equatorium and universal latitude-independent astrolabe by Al-Zarqali (c. AD 1015); the astronomical analog computers of other medieval Muslim astronomers and engineers; and the astronomical clock tower of Su Song (1094) during the Song dynasty. The castle clock, a hydropowered mechanical astronomical clock invented by Ismail al-Jazari in 1206, was the first programmable analog computer.^{[disputed (for: The cited source doesn't support the claim, and the claim is misleading.)  – discuss][12][13][14]} Ramon Llull invented the Lullian Circle: a notional machine for calculating answers to philosophical questions (in this case, to do with Christianity) via logical combinatorics. This idea was taken up by Leibniz centuries later, and is thus one of the founding elements in computing and information science.

Renaissance calculating tools

Scottish mathematician and physicist John Napier discovered that the multiplication and division of numbers could be performed by the addition and subtraction, respectively, of the logarithms of those numbers. While producing the first logarithmic tables, Napier needed to perform many tedious multiplications. It was at this point that he designed his 'Napier's bones', an abacus-like device that greatly simplified calculations that involved multiplication and division.^[d]

Since real numbers can be represented as distances or intervals on a line, the slide rule was invented in the 1620s, shortly after Napier's work, to allow multiplication and division operations to be carried out significantly faster than was previously possible.^[15] Edmund Gunter built a calculating device with a single logarithmic scale at the University of Oxford. His device greatly simplified arithmetic calculations, including multiplication and division. William Oughtred greatly improved this in 1630 with his circular slide rule. He followed this up with the modern slide rule in 1632, essentially a combination of two Gunter rules, held together with the hands. Slide rules were used by generations of engineers and other mathematically involved professional workers, until the invention of the pocket calculator.^[16]

Mechanical calculators

In 1609, Guidobaldo del Monte made a mechanical multiplier to calculate fractions of a degree. Based on a system of four gears, the rotation of an index on one quadrant corresponds to 60 rotations of another index on an opposite quadrant.^[17] Thanks to this machine, errors in the calculation of first, second, third and quarter degrees can be avoided. Guidobaldo is the first to document the use of gears for mechanical calculation.

Wilhelm Schickard, a German polymath, designed a calculating machine in 1623 which combined a mechanized form of Napier's rods with the world's first mechanical adding machine built into the base. Because it made use of a single-tooth gear there were circumstances in which its carry mechanism would jam.^[18] A fire destroyed at least one of the machines in 1624 and it is believed Schickard was too disheartened to build another.

In 1642, while still a teenager, Blaise Pascal started some pioneering work on calculating machines and after three years of effort and 50 prototypes^[19] he invented a mechanical calculator.^[20][21] He built twenty of these machines (called Pascal's calculator or Pascaline) in the following ten years.^[22] Nine Pascalines have survived, most of which are on display in European museums.^[e] A continuing debate exists over whether Schickard or Pascal should be regarded as the "inventor of the mechanical calculator" and the range of issues to be considered is discussed elsewhere.^[23]

Gottfried Wilhelm von Leibniz invented the stepped reckoner and his famous stepped drum mechanism around 1672. He attempted to create a machine that could be used not only for addition and subtraction but would use a moveable carriage to enable multiplication and division. Leibniz once said "It is unworthy of excellent men to lose hours like slaves in the labour of calculation which could safely be relegated to anyone else if machines were used."^[24] However, Leibniz did not incorporate a fully successful carry mechanism. Leibniz also described the binary numeral system,^[25] a central ingredient of all modern computers. However, up to the 1940s, many subsequent designs (including Charles Babbage's machines of 1822 and even ENIAC of 1945) were based on the decimal system.^[f]

Around 1820, Charles Xavier Thomas de Colmar created what would over the rest of the century become the first successful, mass-produced mechanical calculator, the Thomas Arithmometer. It could be used to add and subtract, and with a moveable carriage the operator could also multiply, and divide by a process of long multiplication and long division.^[26] It utilised a stepped drum similar in conception to that invented by Leibniz. Mechanical calculators remained in use until the 1970s.

Punched-card data processing

In 1804, French weaver Joseph Marie Jacquard developed a loom in which the pattern being woven was controlled by a paper tape constructed from punched cards. The paper tape could be changed without altering the mechanical design of the loom. This was a landmark achievement in programmability. His machine was an improvement over similar weaving looms. Punched cards were preceded by punch bands, as in the machine proposed by Basile Bouchon. These bands would inspire information recording for automatic pianos and more recently numerical control machine tools.

In the late 1880s, the American Herman Hollerith invented data storage on punched cards that could then be read by a machine.^[27] To process these punched cards, he invented the tabulator and the keypunch machine. His machines used electromechanical relays and counters.^[28] Hollerith's method was used in the 1890 United States census. That census was processed two years faster than the prior census had been.^[29] Hollerith's company eventually became the core of IBM.

By 1920, electromechanical tabulating machines could add, subtract, and print accumulated totals.^[30] Machine functions were directed by inserting dozens of wire jumpers into removable control panels. When the United States instituted Social Security in 1935, IBM punched-card systems were used to process records of 26 million workers.^[31] Punched cards became ubiquitous in industry and government for accounting and administration.

Leslie Comrie's articles on punched-card methods^[32] and W. J. Eckert's publication of Punched Card Methods in Scientific Computation in 1940, described punched-card techniques sufficiently advanced to solve some differential equations or perform multiplication and division using floating-point representations, all on punched cards and unit record machines.^[33] Such machines were used during World War II for cryptographic statistical processing,^[34] as well as a vast number of administrative uses. The Astronomical Computing Bureau of Columbia University performed astronomical calculations representing the state of the art in computing.^[35][36]

Calculators

By the 20th century, earlier mechanical calculators, cash registers, accounting machines, and so on were redesigned to use electric motors, with gear position as the representation for the state of a variable. The word "computer" was a job title assigned to primarily women who used these calculators to perform mathematical calculations.^[37] By the 1920s, British scientist Lewis Fry Richardson's interest in weather prediction led him to propose human computers and numerical analysis to model the weather; to this day, the most powerful computers on Earth are needed to adequately model its weather using the Navier–Stokes equations.^[38]

Companies like Friden, Marchant Calculator and Monroe made desktop mechanical calculators from the 1930s that could add, subtract, multiply and divide.^[39] In 1948, the Curta was introduced by Austrian inventor Curt Herzstark. It was a small, hand-cranked mechanical calculator and as such, a descendant of Gottfried Leibniz's Stepped Reckoner and Thomas' Arithmometer.

The world's first all-electronic desktop calculator was the British Bell Punch ANITA, released in 1961.^[40][41] It used vacuum tubes, cold-cathode tubes and Dekatrons in its circuits, with 12 cold-cathode "Nixie" tubes for its display. The ANITA sold well since it was the only electronic desktop calculator available, and was silent and quick. The tube technology was superseded in June 1963 by the U.S. manufactured Friden EC-130, which had an all-transistor design, a stack of four 13-digit numbers displayed on a 5-inch (13 cm) CRT, and introduced reverse Polish notation (RPN).

Electromechanical computers

The era of modern computing began with a flurry of development before and during World War II. Most digital computers built in this period were electromechanical – electric switches drove mechanical relays to perform the calculation. These mechanical components had a low operating speed due to their mechanical nature and were eventually superseded by much faster all-electric components, originally using vacuum tubes and later transistors.

The Z2 was one of the earliest examples of an electric operated digital computer built with electromechanical relays and was created by civil engineer Konrad Zuse in 1940 in Germany. It was an improvement on his earlier, mechanical Z1; although it used the same mechanical memory, it replaced the arithmetic and control logic with electrical relay circuits.^[73]

In the same year, electro-mechanical devices called bombes were built by British cryptologists to help decipher German Enigma-machine-encrypted secret messages during World War II. The bombe's initial design was created in 1939 at the UK Government Code and Cypher School at Bletchley Park by Alan Turing,^[74] with an important refinement devised in 1940 by Gordon Welchman.^[75] The engineering design and construction was the work of Harold Keen of the British Tabulating Machine Company. It was a substantial development from a device that had been designed in 1938 by Polish Cipher Bureau cryptologist Marian Rejewski, and known as the "cryptologic bomb" (Polish: "bomba kryptologiczna").

In 1941, Zuse followed his earlier machine up with the Z3,^[73] the world's first working electromechanical programmable, fully automatic digital computer.^[76] The Z3 was built with 2000 relays, implementing a 22-bit word length that operated at a clock frequency of about 5–10 Hz.^[77] Program code and data were stored on punched film. It was similar to modern machines in several respects, pioneering numerous advances such as floating-point numbers. Replacement of the hard-to-implement decimal system (used in Charles Babbage's earlier design) by the simpler binary system meant that Zuse's machines were easier to build and potentially more reliable, given the technologies available at that time.^[78] Despite lacking explicit conditional execution, the Z3 was proven to have been a theoretically Turing-complete machine in 1998 by Raúl Rojas.^[79] In two 1936 patent applications, Zuse also anticipated that machine instructions could be stored in the same storage used for data—the key insight of what became known as the von Neumann architecture, first implemented in 1948 in America in the electromechanical IBM SSEC and in Britain in the fully electronic Manchester Baby.^[80]

Zuse suffered setbacks during World War II when some of his machines were destroyed in the course of Allied bombing campaigns. Apparently his work remained largely unknown to engineers in the UK and US until much later, although at least IBM was aware of it as it financed his post-war startup company in 1946 in return for an option on Zuse's patents.

In 1944, the Harvard Mark I was constructed at IBM's Endicott laboratories.^[81] It was a similar general-purpose electro-mechanical computer to the Z3, but was not quite Turing-complete.

Digital computation

The term digital was first suggested by George Robert Stibitz and refers to where a signal, such as a voltage, is not used to directly represent a value (as it would be in an analog computer), but to encode it. In November 1937, Stibitz, then working at Bell Labs (1930–1941),^[82] completed a relay-based calculator he later dubbed the "Model K" (for "kitchen table", on which he had assembled it), which became the first binary adder.^[83] Typically signals have two states – low (usually representing 0) and high (usually representing 1), but sometimes three-valued logic is used, especially in high-density memory. Modern computers generally use binary logic, but many early machines were decimal computers. In these machines, the basic unit of data was the decimal digit, encoded in one of several schemes, including binary-coded decimal or BCD, bi-quinary, excess-3, and two-out-of-five code.

The mathematical basis of digital computing is Boolean algebra, developed by the British mathematician George Boole in his work The Laws of Thought, published in 1854. His Boolean algebra was further refined in the 1860s by William Jevons and Charles Sanders Peirce, and was first presented systematically by Ernst Schröder and A. N. Whitehead.^[84] In 1879 Gottlob Frege developed the formal approach to logic and proposes the first logic language for logical equations.^[85]

In the 1930s and working independently, American electronic engineer Claude Shannon and Soviet logician Victor Shestakov both showed a one-to-one correspondence between the concepts of Boolean logic and certain electrical circuits, now called logic gates, which are now ubiquitous in digital computers.^[86] They showed that electronic relays and switches can realize the expressions of Boolean algebra.^[87] This thesis essentially founded practical digital circuit design. In addition Shannon's paper gives a correct circuit diagram for a 4 bit digital binary adder.^[88]

Electronic data processing

Purely electronic circuit elements soon replaced their mechanical and electromechanical equivalents, at the same time that digital calculation replaced analog. Machines such as the Z3, the Atanasoff–Berry Computer, the Colossus computers, and the ENIAC were built by hand, using circuits containing relays or valves (vacuum tubes), and often used punched cards or punched paper tape for input and as the main (non-volatile) storage medium.^[89]

Engineer Tommy Flowers joined the telecommunications branch of the General Post Office in 1926. While working at the research station in Dollis Hill in the 1930s, he began to explore the possible use of electronics for the telephone exchange. Experimental equipment that he built in 1934 went into operation 5 years later, converting a portion of the telephone exchange network into an electronic data processing system, using thousands of vacuum tubes.^[61]

In the US, in 1940 Arthur Dickinson (IBM) invented the first digital electronic computer.^[90] This calculating device was fully electronic – control, calculations and output (the first electronic display).^[91] John Vincent Atanasoff and Clifford E. Berry of Iowa State University developed the Atanasoff–Berry Computer (ABC) in 1942,^[92] the first binary electronic digital calculating device.^[93] This design was semi-electronic (electro-mechanical control and electronic calculations), and used about 300 vacuum tubes, with capacitors fixed in a mechanically rotating drum for memory. However, its paper card writer/reader was unreliable and the regenerative drum contact system was mechanical. The machine's special-purpose nature and lack of changeable, stored program distinguish it from modern computers.^[94]

Computers whose logic was primarily built using vacuum tubes are now known as first generation computers.

The electronic programmable computer

During World War II, British codebreakers at Bletchley Park, 40 miles (64 km) north of London, achieved a number of successes at breaking encrypted enemy military communications. The German encryption machine, Enigma, was first attacked with the help of the electro-mechanical bombes.^[95] They ruled out possible Enigma settings by performing chains of logical deductions implemented electrically. Most possibilities led to a contradiction, and the few remaining could be tested by hand.

The Germans also developed a series of teleprinter encryption systems, quite different from Enigma. The Lorenz SZ 40/42 machine was used for high-level Army communications, code-named "Tunny" by the British. The first intercepts of Lorenz messages began in 1941. As part of an attack on Tunny, Max Newman and his colleagues developed the Heath Robinson, a fixed-function machine to aid in code breaking.^[96] Tommy Flowers, a senior engineer at the Post Office Research Station^[97] was recommended to Max Newman by Alan Turing^[98] and spent eleven months from early February 1943 designing and building the more flexible Colossus computer (which superseded the Heath Robinson).^[99][100] After a functional test in December 1943, Colossus was shipped to Bletchley Park, where it was delivered on 18 January 1944^[101] and attacked its first message on 5 February.^[102] By the time Germany surrendered in May 1945, there were ten Colossi working at Bletchley Park.^[103]

Colossus was the world's first electronic digital programmable computer.^[61] It used a large number of valves (vacuum tubes). It had paper-tape input and was capable of being configured to perform a variety of Boolean logical operations on its data,^[104] but it was not Turing-complete. Data input to Colossus was by photoelectric reading of a paper tape transcription of the enciphered intercepted message. This was arranged in a continuous loop so that it could be read and re-read multiple times – there being no internal store for the data. The reading mechanism ran at 5,000 characters per second with the paper tape moving at 40 ft/s (12.2 m/s; 27.3 mph). Colossus Mark 1 contained 1500 thermionic valves (tubes), but Mark 2 with 2400 valves and five processors in parallel, was both 5 times faster and simpler to operate than Mark 1, greatly speeding the decoding process. Mark 2 was designed while Mark 1 was being constructed. Allen Coombs took over leadership of the Colossus Mark 2 project when Tommy Flowers moved on to other projects.^[105] The first Mark 2 Colossus became operational on 1 June 1944, just in time for the Allied Invasion of Normandy on D-Day.

Most of the use of Colossus was in determining the start positions of the Tunny rotors for a message, which was called "wheel setting". Colossus included the first-ever use of shift registers and systolic arrays, enabling five simultaneous tests, each involving up to 100 Boolean calculations. This enabled five different possible start positions to be examined for one transit of the paper tape.^[106] As well as wheel setting some later Colossi included mechanisms intended to help determine pin patterns known as "wheel breaking". Both models were programmable using switches and plug panels in a way their predecessors had not been.

Without the use of these machines, the Allies would have been deprived of the very valuable intelligence that was obtained from reading the vast quantity of enciphered high-level telegraphic messages between the German High Command (OKW) and their army commands throughout occupied Europe. Details of their existence, design, and use were kept secret well into the 1970s. Winston Churchill personally issued an order for their destruction into pieces no larger than a man's hand, to keep secret that the British were capable of cracking Lorenz SZ cyphers (from German rotor stream cipher machines) during the oncoming Cold War. Two of the machines were transferred to the newly formed GCHQ and the others were destroyed. As a result, the machines were not included in many histories of computing.^[g] A reconstructed working copy of one of the Colossus machines is now on display at Bletchley Park.

The ENIAC (Electronic Numerical Integrator and Computer) was the first electronic programmable computer built in the US. Although the ENIAC used similar technology to the Colossi, it was much faster and more flexible and was Turing-complete. Like the Colossi, a "program" on the ENIAC was defined by the states of its patch cables and switches, a far cry from the stored-program electronic machines that came later. Once a program was ready to be run, it had to be mechanically set into the machine with manual resetting of plugs and switches. The programmers of the ENIAC were women who had been trained as mathematicians.^[108]

It combined the high speed of electronics with the ability to be programmed for many complex problems. It could add or subtract 5000 times a second, a thousand times faster than any other machine. It also had modules to multiply, divide, and square root. High-speed memory was limited to 20 words (equivalent to about 80 bytes). Built under the direction of John Mauchly and J. Presper Eckert at the University of Pennsylvania, ENIAC's development and construction lasted from 1943 to full operation at the end of 1945. The machine was huge, weighing 30 tons, using 200 kilowatts of electric power and contained over 18,000 vacuum tubes, 1,500 relays, and hundreds of thousands of resistors, capacitors, and inductors.^[109] One of its major engineering feats was to minimize the effects of tube burnout, which was a common problem in machine reliability at that time. The machine was in almost constant use for the next ten years.

Manchester Baby

The Manchester Baby (Small Scale Experimental Machine, SSEM) was the world's first electronic stored-program computer. It was built at the Victoria University of Manchester by Frederic C. Williams, Tom Kilburn and Geoff Tootill, and ran its first program on 21 June 1948.^[114]

The machine was not intended to be a practical computer, but was instead designed as a testbed for the Williams tube, the first random-access digital storage device.^[115] Invented by Freddie Williams and Tom Kilburn^[116][117] at the University of Manchester in 1946 and 1947, it was a cathode-ray tube that used an effect called secondary emission to temporarily store electronic binary data, and was used successfully in several early computers.

Described as small and primitive in a 1998 retrospective, the Baby was the first working machine to contain all of the elements essential to a modern electronic computer.^[118] As soon as it had demonstrated the feasibility of its design, a project was initiated at the university to develop the design into a more usable computer, the Manchester Mark 1. The Mark 1 in turn quickly became the prototype for the Ferranti Mark 1, the world's first commercially available general-purpose computer.^[119]

The Baby had a 32-bit word length and a memory of 32 words. As it was designed to be the simplest possible stored-program computer, the only arithmetic operations implemented in hardware were subtraction and negation; other arithmetic operations were implemented in software. The first of three programs written for the machine found the highest proper divisor of 2¹⁸ (262,144), a calculation that was known would take a long time to run—and so prove the computer's reliability—by testing every integer from 2¹⁸ − 1 downwards, as division was implemented by repeated subtraction of the divisor. The program consisted of 17 instructions and ran for 52 minutes before producing the correct answer of 131,072, after the Baby had performed 3.5 million operations (for an effective CPU speed of 1.1 kIPS). The successive approximations to the answer were displayed as a pattern of dots on the output CRT which mirrored the pattern held on the Williams tube used for storage.

Manchester Mark 1

The SSEM led to the development of the Manchester Mark 1 at the University of Manchester.^[120] Work began in August 1948, and the first version was operational by April 1949; a program written to search for Mersenne primes ran error-free for nine hours on the night of 16/17 June 1949. The machine's successful operation was widely reported in the British press, which used the phrase "electronic brain" in describing it to their readers.

The computer is especially historically significant because of its pioneering inclusion of index registers, an innovation which made it easier for a program to read sequentially through an array of words in memory. Thirty-four patents resulted from the machine's development, and many of the ideas behind its design were incorporated in subsequent commercial products such as the IBM 701 and 702 as well as the Ferranti Mark 1. The chief designers, Frederic C. Williams and Tom Kilburn, concluded from their experiences with the Mark 1 that computers would be used more in scientific roles than in pure mathematics. In 1951 they started development work on Meg, the Mark 1's successor, which would include a floating-point unit.

EDSAC

The other contender for being the first recognizably modern digital stored-program computer^[121] was the EDSAC,^[122] designed and constructed by Maurice Wilkes and his team at the University of Cambridge Mathematical Laboratory in England at the University of Cambridge in 1949. The machine was inspired by John von Neumann's seminal First Draft of a Report on the EDVAC and was one of the first usefully operational electronic digital stored-program computers.^[h]

EDSAC ran its first programs on 6 May 1949, when it calculated a table of squares^[125] and a list of prime numbers.The EDSAC also served as the basis for the first commercially applied computer, the LEO I, used by food manufacturing company J. Lyons & Co. Ltd. EDSAC 1 was finally shut down on 11 July 1958, having been superseded by EDSAC 2 which stayed in use until 1965.^[126]

The "brain" [computer] may one day come down to our level [of the common people] and help with our income-tax and book-keeping calculations. But this is speculation and there is no sign of it so far.

EDVAC

ENIAC inventors John Mauchly and J. Presper Eckert proposed the EDVAC's construction in August 1944, and design work for the EDVAC commenced at the University of Pennsylvania's Moore School of Electrical Engineering, before the ENIAC was fully operational. The design implemented a number of important architectural and logical improvements conceived during the ENIAC's construction, and a high-speed serial-access memory.^[112] However, Eckert and Mauchly left the project and its construction floundered.

It was finally delivered to the U.S. Army's Ballistics Research Laboratory at the Aberdeen Proving Ground in August 1949, but due to a number of problems, the computer only began operation in 1951, and then only on a limited basis.

Commercial computers

The first commercial electronic computer was the Ferranti Mark 1, built by Ferranti and delivered to the University of Manchester in February 1951. It was based on the Manchester Mark 1. The main improvements over the Manchester Mark 1 were in the size of the primary storage (using random access Williams tubes), secondary storage (using a magnetic drum), a faster multiplier, and additional instructions. The basic cycle time was 1.2 milliseconds, and a multiplication could be completed in about 2.16 milliseconds. The multiplier used almost a quarter of the machine's 4,050 vacuum tubes (valves).^[128] A second machine was purchased by the University of Toronto, before the design was revised into the Mark 1 Star. At least seven of these later machines were delivered between 1953 and 1957, one of them to Shell labs in Amsterdam.^[129]

In October 1947, the directors of J. Lyons & Company, a British catering company famous for its teashops but with strong interests in new office management techniques, decided to take an active role in promoting the commercial development of computers. The LEO I computer (Lyons Electronic Office) became operational in April 1951^[130] and ran the world's first regular routine office computer job. On 17 November 1951, the J. Lyons company began weekly operation of a bakery valuations job on the LEO – the first business application to go live on a stored-program computer.^[i]

In June 1951, the UNIVAC I (Universal Automatic Computer) was delivered to the U.S. Census Bureau. Remington Rand eventually sold 46 machines at more than US$1 million each ($12.4 million as of 2025).^[131] UNIVAC was the first "mass-produced" computer. It used 5,200 vacuum tubes and consumed 125 kW of power. Its primary storage was serial-access mercury delay lines capable of storing 1,000 words of 11 decimal digits plus sign (72-bit words).

In 1952, Compagnie des Machines Bull released the Gamma 3 computer, which became a large success in Europe, eventually selling more than 1,200 units, and the first computer produced in more than 1,000 units.^[132] The Gamma 3 had innovative features for its time including a dual-mode, software switchable, BCD and binary ALU, as well as a hardwired floating-point library for scientific computing.^[132] In its E.T configuration, the Gamma 3 drum memory could fit about 50,000 instructions for a capacity of 16,384 words (around 100 kB), a large amount for the time.^[132]

Compared to the UNIVAC, IBM introduced a smaller, more affordable computer in 1954 that proved very popular.^[j][134] The IBM 650 weighed over 900 kg, the attached power supply weighed around 1350 kg and both were held in separate cabinets of roughly 1.5 × 0.9 × 1.8 m. The system cost US$500,000^[135] ($5.99 million as of 2025) or could be leased for US$3,500 a month ($40,000 as of 2025).^[131] Its drum memory was originally 2,000 ten-digit words, later expanded to 4,000 words. Memory limitations such as this were to dominate programming for decades afterward. The program instructions were fetched from the spinning drum as the code ran. Efficient execution using drum memory was provided by a combination of hardware architecture – the instruction format included the address of the next instruction – and software: the Symbolic Optimal Assembly Program, SOAP,^[136] assigned instructions to the optimal addresses (to the extent possible by static analysis of the source program). Thus many instructions were, when needed, located in the next row of the drum to be read and additional wait time for drum rotation was reduced.

Microprogramming

In 1951, British scientist Maurice Wilkes developed the concept of microprogramming from the realisation that the central processing unit of a computer could be controlled by a miniature, highly specialized computer program in high-speed ROM. Microprogramming allows the base instruction set to be defined or extended by built-in programs (now called firmware or microcode).^[137] This concept greatly simplified CPU development. He first described this at the University of Manchester Computer Inaugural Conference in 1951, then published in expanded form in IEEE Spectrum in 1955.^{[citation needed]}

It was widely used in the CPUs and floating-point units of mainframe and other computers; it was implemented for the first time in EDSAC 2,^[138] which also used multiple identical "bit slices" to simplify design. Interchangeable, replaceable tube assemblies were used for each bit of the processor.

Transistor peripherals

Transistorized electronics improved not only the CPU (Central Processing Unit), but also the peripheral devices. The second generation disk data storage units were able to store tens of millions of letters and digits. Next to the fixed disk storage units, connected to the CPU via high-speed data transmission, were removable disk data storage units. A removable disk pack can be easily exchanged with another pack in a few seconds. Even if the removable disks' capacity is smaller than fixed disks, their interchangeability guarantees a nearly unlimited quantity of data close at hand. Magnetic tape provided archival capability for this data, at a lower cost than disk.

Many second-generation CPUs delegated peripheral device communications to a secondary processor. For example, while the communication processor controlled card reading and punching, the main CPU executed calculations and binary branch instructions. One databus would bear data between the main CPU and core memory at the CPU's fetch-execute cycle rate, and other databusses would typically serve the peripheral devices. On the PDP-1, the core memory's cycle time was 5 microseconds; consequently most arithmetic instructions took 10 microseconds (100,000 operations per second) because most operations took at least two memory cycles; one for the instruction, one for the operand data fetch.

During the second generation remote terminal units (often in the form of Teleprinters like a Friden Flexowriter) saw greatly increased use.^[k] Telephone connections provided sufficient speed for early remote terminals and allowed hundreds of kilometers separation between remote-terminals and the computing center. Eventually these stand-alone computer networks would be generalized into an interconnected network of networks—the Internet.^[l]

Transistor supercomputers

The early 1960s saw the advent of supercomputing. The Atlas was a joint development between the University of Manchester, Ferranti, and Plessey, and was first installed at Manchester University and officially commissioned in 1962 as one of the world's first supercomputers – considered to be the most powerful computer in the world at that time.^[155] It was said that whenever Atlas went offline half of the United Kingdom's computer capacity was lost.^[156] It was a second-generation machine, using discrete germanium transistors. Atlas also pioneered the Atlas Supervisor, "considered by many to be the first recognisable modern operating system".^[157]

In the US, a series of computers at Control Data Corporation (CDC) were designed by Seymour Cray to use innovative designs and parallelism to achieve superior computational peak performance.^[158] The CDC 6600, released in 1964, is generally considered the first supercomputer.^[159][160] The CDC 6600 outperformed its predecessor, the IBM 7030 Stretch, by about a factor of 3. With performance of about 1 megaFLOPS, the CDC 6600 was the world's fastest computer from 1964 to 1969, when it relinquished that status to its successor, the CDC 7600.