STARS:Early Popular Computers, 1950 - 1970
Author: Robert Garner
The 1951 introduction of the large-scale UNIVAC computer from Remington Rand began a three-decade transition from over half a century of data processing on punched-card equipment to the widespread use of stored-program computers. While large firms profitably employed early large-scale computers, most businesses could not afford them. The transition to computers gained momentum in the mid-1950s with introduction of the magnetic-drum IBM 650 and the magnetic-disk IBM 305, and again in the 1960s with the magnetic-core and transistorized IBM 1401. With deliveries tripling every year, computer revenues exceeded those of punched-card equipment by 1962.
|1947||Bell Labs invents the transistor|
|1948||IBM delivers vacuum-tube 604 Electronic Calculating Punch|
|1949||IBM delivers 604-based Card-Programmed Electronic Calculator (CPC)|
|1951||Remington Rand delivers vacuum-tube UNIVAC computer|
|1952||Bell labs introduces the alloy-junction transistor|
|1954||IBM delivers vacuum-tube 650 Magnetic Drum Data Processing System|
|1955||IBM delivers magnetic-core vacuum-tube SAGE air-defense computer|
|1956||IBM delivers vacuum-tube 305 RAMAC, first computer with magnetic disks|
|1958||Philco delivers S-2000 Transistorized Automatic Computer|
|1959||IBM announces transistorized 1401 Data Processing System|
|1960||IBM 650 shipments exceed 1,000|
|1962||Computer revenues overtake those of punched-card equipment|
|1964||IBM 1401 shipments exceed 10,000|
|1973||IBM System/360 Models 20 and 30 shipments exceed 20,000|
|1974||IBM System/3 shipments exceed 25,000|
Over the past six decades, the raw processing power of computers has increased by six orders of magnitude while cost has dropped by three orders of magnitude. Computers have become ubiquitous, invisibly embedded in everyday items such as mobile phones, cameras, automobiles and entertainment systems.
By comparison, in the early-1950s, only the largest firms and institutions could afford a computer. Leasing for about $200,000 per month in inflation-adjusted 2010 dollars, a large-scale computer occupied roughly a hundred square meters, used thousands of vacuum tubes, consumed over a hundred kilowatts, and needed a full-time operating and maintenance staff.
In order to broaden the market, vendors endeavored to offer computers to replace the punched-card equipment widely used by businesses and institutions for over half a century. In this essay, we highlight the most popular and widely used computers from the early-1950s through the 1960s. These computers, manufactured in substantial quantities relative to their time, spearheaded the transition from punched-card equipment to stored-program computing.
Beginning with the invention of the tabulator, keypunch and sorter in the late-1880s, punched-card equipment became progressively more versatile and widely employed in the automation of business tasks such as payroll, billing, inventory, and receivables. By 1950, vendors such as IBM and Remington Rand offered a variety of punched-card machines: keypunches, accounting machines/tabulators (printers), sorters, collators, calculators, interpreters and reproducers. IBM’s card with rectangular holes, introduced in 1928, held up to 80 characters while Remington Rand’s card with circular holes, introduced two years later, held up to 90 characters.
Punched-card equipment processed decks of cards, each machine tailored for a particular task by a removable plugboard control panel. By manually plugging jumper wires, users “programmed” calculation or control steps and card movement from input card hoppers to output card stackers. Punched-card machines were electromechanical, using electrical circuits, relays and electromagnets for control and electric motors for moving punched cards and paper via pulleys, rollers, gears, levers and blades.
Small business firms in the 1950s automated their data processing tasks by leasing an assortment of punched-card machines for about $2,500 per month ($20,000 inflation adjusted). They consumed millions of punched cards per year, which often cost more than the equipment rental itself.
First introduced in 1935, the electromechanical multiplying punch supplemented accounting machines and was used for scientific and engineering calculations. In 1946, IBM announced its 603 Electronic Multiplier, the first vacuum-tube calculator; and two years later, it introduced the more capable 604 Electronic Calculating Punch. Renting for $550 per month ($5,000 inflation adjusted), the 604 could multiply or divide about a dozen numbers per second. Not today’s hand held calculator, it had 1,400 vacuum tubes, weighed 630 kilograms and consumed 8,000 watts.
In 1948, engineers at Northrop Aircraft in California prompted IBM to develop its Card-Programmed Calculator (CPC). While a stored-program computer executes instructions and data from main memory, the CPC used an augmented 604 to execute instructions from punched cards. About 4,500 IBM 604s and 700 CPCs were installed worldwide by the mid-1950s. Users affectionately referred to the CPC as “a poor man’s ENIAC.”
Early large-scale commercial computers
In 1951, the US Census Bureau took delivery of Remington Rand’s UNIVAC stored-program computer. It rented for $16,000 per month ($130,000 inflation adjusted), had a 1000-word memory with 12 characters per word and offered magnetic tape drives. A total of 43 UNIVACs were ultimately delivered to businesses and institutions during the 1950s.
In 1953, IBM delivered its 701 Electronic Data Processing Machine, a large-scale computer for scientific and engineering applications, including national defense calculations. Two years later, IBM introduced its 702 Data Processing System for businesses.
Other US manufacturers of large-scale commercial computers during the 1950s included Burroughs, Datamatic, Engineering Research Associates (ERA), GE, NCR, RCA, and Philco. European manufacturers included Bull, Elliott, Electrologica, EMI, English Electric, Facit, Ferranti, ICT, Lyons, Olivetti, SEA, Siemens, and Zuse. By the close of the 1950s, a dozen US vendors had delivered approximately 450 large-scale computers, of which 350 were from IBM.
Large firms found that, despite their high costs, large-scale computers could pay for themselves by consolidating accounting, clerical and record keeping services across different departments. Also, by employing mathematical algorithms from the field of operations research, manufacturing firms could optimize their distribution of goods and materials. Companies and institutions proudly showcased computers in iconic glass-fronted rooms.
Reinforcing the preeminence of large-scale computers, the industry consultant Herbert Grosch asserted that computational capability was proportional to the square of a computer’s cost: twice the cost should yield four times the capability. Nevertheless, despite “Grosch’s Law,” firms sought less costly computers that they could wholly control for faster programming development cycles and freedom from the interruptions of a shared service.
Early medium-scale computers
By the mid-1950s, over a dozen intrepid vendors endeavored to deliver inexpensive vacuum-tube computers. Limited in their capabilities, none were shipped in quantities of more than a hundred or so.
While IBM developed large-scale computers in its laboratory in Poughkeepsie, New York, engineers in its upstate Endicott laboratory designed a medium-scale computer based on a rotating magnetic drum for memory. Delivered in 1954, the IBM 650 Magnetic Drum Data Processing System targeted scientific, engineering and business applications [Figure 1]. Like the large-scale computers, it supported magnetic-tape drives and also offered a scientific floating-point arithmetic option. A usable entry configuration leased for $4,000 per month ($32,000 inflation adjusted). Using vacuum tubes for control circuits, the 650 offered 2,000 words (of ten digits or five alphanumeric characters each) on a 10-cm diameter drum spinning at 12,500 revolutions per minute.
A few years later at IBM’s new laboratory in San Jose, California, engineers pioneered a computer based on the world’s first moving-head magnetic disk drive. Delivered in 1956, the IBM 305 RAMAC (Random Access Method of Accounting and Control) targeted business applications such as inventory, billing, accounts receivable and transaction processing. The entry configuration leased for an attractive $3,200 per month ($25,000 inflation adjusted). The 305 used vacuum tubes and a plugboard control panel and could execute up to 200 instructions from a small magnetic drum memory. The RAMAC stored up to 5 million characters on a stack of fifty 61-cm diameter disks spinning at 1,200 revolutions per minute.
By 1956, a computer census reported that three-quarter of the world’s 1,100 computer installations were IBM 650s, making it by far the most popular computer at the time. Its numbers later peaked at 1,800 installations. The IBM 305 RAMAC was nearly as successful, with over 1,000 delivered before its withdrawal in 1961.
Although the IBM 650 and 305 cost less than one fifth that of a large-scale computer, many small businesses still found them too expensive, with a monthly rental exceeding that of their typical assortment of punched-card machines.
During this time, computer designs normally targetted either scientific or business applications, but not both. In general, scientific and engineering applications benefitted from the fastest possible arithmetic calculations, whereas data processing applications did not benefit from processing speeds faster than the input/output data rates of peripheral devices. A design focused on business data processing applications, together with less expensive circuits, was required before the economical needs of the small-business punched-card marketplace could be met. This suggested a cost target close to the $2,500 monthly rental for a small business’s typical assortment of punched-card equipment.
Magnetic-core arrays for memory
Early computers used ingenious but challenging memory technologies. Acoustic delay lines in the UNIVAC and rotating magnetic drums in ERA and IBM computers yielded slow and non-uniform access times relative to the speed of vacuum-tube circuits. The random-access electrostatic cathode-ray-tube memories used in the IBM 701 and 702 were fast enough but not sufficiently reliable.
In the early-1950s, MIT’s Jay Forrester, working on the government-funded Whirlwind computer project, pioneered a new type of memory that used arrays of tiny doughnut-shaped magnetic cores threaded by control wires [Figure 2]. Clockwise or counterclockwise magnetization of a core represented a logical 0 or 1. By late 1952, Forrester had demonstrated a 16×16 core-memory array and IBM independently prototyped an 80×12 core array that could buffer data from a punched card.
In late 1952, IBM contracted with MIT for a new large-scale computer for SAGE, an enormous real-time surveillance and control system to defend US airspace against the threat of Soviet Union bombers. Delivered in 1955, IBM’s first SAGE computer used a large memory stack comprised of thirty-six 64×64 magnetic-core arrays.
Up until this time, magnetic-core arrays were predominately hand assembled by women who deftly threaded hypodermic needles through selected cores, requiring forty hours per array. Because the SAGE installations would need tens of millions of cores, IBM built a core threading machine and a facility that could fabricate and test half a million cores per day.
By the late 1950s, nearly all computers were using magnetic-core memory arrays. With access times under ten microseconds, they were much faster than acoustic serial delay lines and rotating drums – but not cheaper. At a monthly rental of about two cents per bit, 4,000 characters of memory rented for about $640 a month, comparable in cost to an accounting machine. (If those prices still prevailed today – inflation adjusted to $1 per byte per month – the rental for a 4-gigabyte memory would be 4 billion dollars per month!)
Transistors for logic circuits
In 1947, researchers at Bell Labs invented the transistor, based on the semiconductor germanium. Several years later, they introduced a more reliable and easier-to-manufacture device, the alloy-junction transistor. Initially more expensive, less reliable and slower than vacuum tubes, transistors were nevertheless predicted to become a hundred times cheaper, more reliable, and faster.
By the mid-1950s, universities and research institutions had prototyped experimental transistorized computers. In 1954, IBM demonstrated a fully transistorized adaptation of their 604 calculator which, in place of 1,400 vacuum tubes, used 2,200 alloy-junction transistors, magnetic-core memory, and 95% less power in half the space. In late 1957, IBM’s CEO Thomas J. Watson, Jr. banned the use of vacuum tubes in machine developments, expediting their conversion to transistors. At year’s end, IBM announced its 608 Electronic Calculator, the first commercial transistorized calculator.
In 1956, in anticipation of mass-produced transistorized computers, IBM chartered a small team in Poughkeepsie to automate the assembly of transistors. By 1959, a fully automated production line was operational and capable of yielding hundreds of alloy-junction transistors per hour [Figure 3]. However, to the dismay of its creators, IBM decided not to manufacture transistors and transferred its automated production facility to Texas Instruments to supply IBM and others.
During this time, IBM also developed its Standard Modular System (SMS) for the packaging of discrete solid-state transistors and circuit components, a more compact successor to its vacuum-tube modules. SMS printed-circuit cards implemented basic computer circuits, such as logic gates and single bit latches [Figure 4]. In 1959, IBM began mass producing SMS cards and wire-wrapped backplanes for its large-scale 7090 and Stretch supercomputers.
Early solid-state computers
With magnetic-core memory and transistors in widespread production, the 1960s witnessed a dramatic increase in the number of computers, with deliveries tripling every year. Table 1 lists the 23 earliest commercial computers delivered with magnetic-core memory and transistor circuits through the end of 1960.
Philco was one of the first out in 1958 with its all solid-state large-scale scientific S-2000 Transistorized Automatic Computer. Using its innovative high-speed surface-barrier transistor, it competed with IBM’s later large-scale scientific 7090 Data Processing System. The Electrologica company in the Netherlands also delivered its X1 medium-scale scientific solid-state computer that same year.
Although the IBM 1401 Data Processing System was not the earliest solid-state commercial computer, it rapidly became the world’s most widely used computer through the 1960s [Figure 5]. The story behind the 1401’s success is explained in part because it targeted the small-business punched-card user marketplace while leveraging IBM’s mass-production capabilities and worldwide customer support organization, as described next in a case study of its product development.
References of Historical Significance
References for Further Reading
About the Author(s)
Robert Garner’s Silicon Valley career spans 33 years of management and engineering at Xerox Systems Development Division (STAR workstation) and Palo Alto Research Center (PARC), Sun Microsystems (SPARC architecture, microprocessors, and Sun-4 workstation), Brocade Communications, and IBM Research. In 2004, he volunteered at the Computer History Museum to lead the restoration of an IBM 1401; and in 2009, he organized the “50th Anniversary of the Legendary IBM 1401” event at the Computer History Museum.