Meteorologist, oceanographer, Python solvem probler, experimenter, Raspberry Pi tinkerer, geek dad, manager, space enthusiast, ham radio lover, science columnist, ADHDer, synthetic intellect curious, linux admin cosplayer.

Era 1 and Era 2 Breakthroughs and Failures: Vacuum Tubes and Discrete Transistors (1946-1968)

Research notes for expanding “From Tubes to GPUs”. Each item is a “moment” with date, machine/project, named people, the breakthrough/failure (2-4 sentences), the consequence, and the citation.

A scope note on dates: the user’s prompt mentioned “1958-08-29” for the Soviet bomb context. The first Soviet atomic device test (RDS-1, “Joe-1”, “First Lightning”) was actually 29 August 1949 in Kazakhstan, not 1958. That 1949 detonation is what triggered the U.S. air defense panic, the founding of the Valley Committee in January 1950, the Project Charles report of 1951, the rescue of Whirlwind from cancellation in March 1950, and the eventual SAGE programme. Items below use 1949.

Section 1: Memory Technology Failures (1946-1953)

Mercury Delay Lines: the Sequential-Access Tax (1944-1953)

  • Date: Patent applied 31 October 1947 (Eckert and Mauchly); first production use UNIVAC I (1951).
  • Machines: EDVAC, EDSAC (May 1949), BINAC (1949), UNIVAC I (1951), CSIRAC (1949).
  • People: William Shockley (Bell Labs, 1942 first concept); J. Presper Eckert (MIT, 1944 development); John Mauchly.
  • Failure: Mercury delay lines were acoustic memory: pulses propagated through a sealed tube of liquid mercury at 1450 m/s, looping back to the input end every millisecond or so. The technology had three crippling defects. It was strictly sequential access – if you wanted bit 327 of a 1024-bit line, you waited for it to come around. Mercury was toxic, the tubes were heavy, the apparatus was mechanically fragile. And the speed of sound in mercury shifted with temperature – the apparatus had to be heated to a precise 40 degrees Celsius and held there.
  • Consequence: Sequential-access memory made it nearly impossible to implement efficient inner loops over arrays. UNIVAC I (1951) needed seven large delay-line memory units for its thousand-word store. The technology was abandoned by the mid-1950s as core memory came in; even before that, von Neumann’s IAS architecture had committed itself to random-access electrostatic storage instead. A scientific computer with a delay-line memory was, in retrospect, a computer that could not run a numerical weather model.
  • Punch: The first generation of stored-program computers had memory you could not address randomly. You waited for your bit to come around the loop.
  • Source: Wikipedia - Mercury delay line

Williams Tubes: 15-Minute MTBF and a Daily Recalibration (1946-1955)

  • Date: British patents filed 11 December 1946 and 2 October 1947 (F. C. Williams and Tom Kilburn at Manchester); first production use Manchester Baby (1948); abandoned for new builds by ~1955.
  • Machines: Manchester Baby (1948), Ferranti Mark 1, IAS Machine (1952), IBM 701 (1952), UNIVAC 1103, SWAC, Whirlwind I (1951-1953).
  • People: Frederick C. Williams; Tom Kilburn; James Pomerene at IAS.
  • Failure: Williams tubes used a modified cathode-ray tube as random-access memory: an electron beam deposited a charge pattern on the inside of the phosphor screen, and a sense plate against the outside of the glass detected the bits. The read was destructive – the electron beam erased what it had detected, requiring an immediate rewrite. The charge pattern decayed in milliseconds and had to be refreshed continuously. The screens drifted with use and aging. Most installations had to be hand-tuned daily by technicians. The IBM 701 in production ran with a mean time between memory faults of about fifteen minutes.
  • Consequence: Whirlwind I in 1951-1953 ran on similar electrostatic-tube memory. The project reportedly spent on the order of $32,000 per month replacing failing tubes – close to 40% of the project’s running costs going to memory tubes alone. The forecast model that JNWPU tried to run on its IBM 701 in 1955 frequently failed mid-computation because a Williams tube fault would corrupt the run; a 24-hour forecast usually took longer than the 30-minute MTBF, which meant the model had to be restarted. This is one of the principal reasons the first JNWPU forecasts (May 1955 onward) were “very disappointing and not usable by forecasters” (NCEP/EMC retrospective).
  • Punch: The first commercial scientific computer in the United States (the IBM 701) had a memory that crashed every fifteen minutes.
  • Source: Wikipedia - Williams tube; Wikipedia - IBM 701

Magnetic Drum Memory: Slow, Cheap, and Outcompeted (1932-1970s)

  • Date: Invented in Austria by Gustav Tauschek in 1932; commercial peak 1950s; replaced as primary memory by ~1965; manufacturing ceased in the 1970s.
  • Machines: Atanasoff-Berry Computer (1942, 3000 bits); IBM 650 (1954, mass-market drum machine); ERA 1101/UNIVAC 1101 (1950); Bendix G-15; Librascope LGP-30 (1956, the machine on which Lorenz discovered chaos).
  • People: Gustav Tauschek (1932 inventor); Andrew Booth at Birkbeck (1947 British implementation).
  • Failure: Drums were a rotating cylinder coated in magnetic material with stationary read/write heads. They were cheap, dense, and non-volatile, but strictly sequential within a track and access times averaged half a revolution – typically several milliseconds. A computer with drum memory could not realistically execute random-access scientific code at scale. The IBM 650 (1954, ~2000 units sold) had drum memory; the LGP-30 (1956, ~500 units) had drum memory; both were inherently slow.
  • Consequence: When the IBM 704 with magnetic core memory shipped in late 1955, drum-based scientific computing was over for the high end. The IBM 650 sold nearly 2000 units between 1954 and 1962 and was “the most popular computer of the 1950s,” but the 704 outsold it on a per-dollar performance basis as soon as it was available. Drums survived as secondary storage (disk’s predecessor) into the 1960s; as primary memory they were dead by 1958. The collapse of drum memory as a competitive technology happened in approximately 1955-1960.
  • Punch: Edward Lorenz discovered deterministic chaos on a drum-memory machine that did 60 operations per second. By the time he published in 1963, drum memory as a scientific-computing technology was dead.
  • Source: Wikipedia - Drum memory; Wikipedia - IBM 650

Section 2: The Whirlwind/SAGE Acceleration Story (1949-1958)

29 August 1949: Joe-1 and the Air Defense Panic

  • Date: 29 August 1949 (Soviet RDS-1 atomic test in Kazakhstan); ADSEC formed under General Vandenberg in December 1949; first ADSEC meeting 20 January 1950.
  • People: George Valley (MIT physics professor, ADSEC chair); General Hoyt Vandenberg (Air Force Chief of Staff); General Muir Fairchild; Jay Forrester; Jerome Wiesner.
  • Breakthrough: The Soviet bomb test, a full year ahead of U.S. intelligence estimates, made continental air defense an immediate priority. In November 1949 Valley wrote to von Karman; on 15 December Vandenberg’s deputy Fairchild had organized an eight-scientist committee. The Air Defense Systems Engineering Committee (ADSEC, the “Valley Committee”) began weekly meetings on 20 January 1950. Within the same window, Forrester at MIT was running Whirlwind on Office of Naval Research money, six months from cancellation – the project had spent over $1.9 million by 1948 and was eating 20% of ONR’s research budget.
  • Consequence: A lunch meeting at the MIT Faculty Club in late January 1950 between Valley, Forrester, and Wiesner started the rescue. By March 1950 MIT was negotiating with ONR to transfer Whirlwind to a new sponsor. The Air Force convened Project Charles (formal letter contract 30 January 1951), recommending in October 1951 that MIT host a permanent Air Force-funded laboratory dedicated to air defense. MIT Lincoln Laboratory was founded in 1951 in Lexington. Whirlwind moved from a research curiosity to the centrepiece of U.S. continental air defense. A flight simulator that had spent 1944-1950 looking for a customer found one in the Soviet bomb.
  • Punch: Whirlwind was six months from cancellation in early 1950 when Stalin saved it.
  • Source: Wikipedia - SAGE; MIT Lincoln Laboratory SAGE history; historyofcomputercommunications.info - Real-Time Computing – The SAGE Project

8 August 1953: Magnetic Core Memory at Whirlwind I

  • Date: 8 August 1953 (first core memory bank installed at Whirlwind, replacing one of two Williams-tube banks); both banks core by autumn 1954.
  • People: Jay Forrester (architect, MIT); William N. Papian (the graduate student who did the materials science and circuitry); An Wang (Harvard, the parallel inventor of destructive-read-with-immediate-rewrite); Robert Everett.
  • Breakthrough: After four years of development from Forrester’s 13 June 1949 notebook entry, Whirlwind got a 32-by-32-by-16 core array of 16,384 bits. Memory access dropped from ~25 microseconds (Williams tubes) to ~9 microseconds. Throughput approximately doubled, from 20,000 to 40,000 single-address operations per second. Most importantly: maintenance time on the memory dropped from four hours per day to two hours per week – a 50- to 100-fold improvement in mean-time-between-failures (Papian himself, the most-cited figure).
  • Consequence: The $32,000-per-month tube-replacement bill simply vanished from MIT’s books. Memory had been Whirlwind’s fragility; now it was its reliability. The first generation of cores installed in August 1953 outlasted Whirlwind itself; the same banks were operating in May 1959 when the machine was decommissioned. Core memory then powered every weather and climate machine for the next 25 years: IBM 704 (Phillips’s GCM); IBM 7090 and 7094 (early GFDL models); CDC 1604 and 6600 and 7600 (NCAR); UNIVAC 1108 (Manabe-Bryan ocean coupling); DEC VAX 11/780 (Cane-Zebiak ENSO).
  • Punch: The single most important architectural step in 25 years of computational meteorology was the 8 August 1953 swap of CRTs for ferrite donuts at MIT.
  • Source: Computer History Museum - 1953: Whirlwind computer debuts core memory; existing Post 28 (Memory in Every Forecast) is the canonical narrative.

26 June 1958: First SAGE Direction Center Goes Operational at McGuire AFB

  • Date: 26 June 1958 (DC-01 at McGuire AFB, New Jersey).
  • Machines: IBM AN/FSQ-7 Combat Direction Central – the largest computer ever built.
  • People: George Valley (ADSEC chair); Jay Forrester (Lincoln Lab); Robert Everett (Lincoln Lab and later MITRE); IBM as prime contractor (selected October 1952).
  • Breakthrough/Verification: Each AN/FSQ-7 contained 49,000 vacuum tubes in the computers (60,000 if peripherals are counted), weighed 250 tons (sometimes cited as 250-275), occupied 2,000 square metres of floor space (roughly half an acre), drew up to 3 megawatts of power, and ran at about 75,000 instructions per second. Each direction center had two of these machines (one active, one hot spare). The Air Force eventually deployed 24 SAGE direction centers plus three combat centers.
  • Consequence: The total program cost across the deployment cycle is most often given as $8 to $12 billion in then-year currency. The Manhattan Project cost $2 billion. SAGE was four to six times the Manhattan Project, by cost. IBM’s production contract alone delivered 56 SAGE computers for approximately $0.5 billion – about $18 million per AN/FSQ-7 pair. The system was technologically obsolete by the time it became fully operational (the threat had shifted from bombers to ballistic missiles), but the network remained in continuous service until 1983-1984 (last AN/FSQ-7 demolished at Luke AFB, February 1984). SAGE’s most consequential legacy: it forced IBM to scale up to industrial-strength production of magnetic-core memory and digital systems, paying for the infrastructure that IBM would then use to dominate commercial mainframe computing for the next 25 years. The Defense Department paid for IBM’s manufacturing capability.
  • Punch: The largest computer ever built ran on 49,000 vacuum tubes in two-machine pairs across 24 sites, cost four to six times the Manhattan Project, and was technologically obsolete the day it went operational.
  • Source: Wikipedia - AN/FSQ-7; Wikipedia - SAGE; IBM History - SAGE

Section 3: The IBM 701 Era (1952-1955)

21 May 1952: IBM 701 Defense Calculator Announced

  • Date: 21 May 1952 (announcement); first installation 590 Madison Avenue, late December 1952; Los Alamos delivery early April 1953; dedication luncheon 7 April 1953 with J. Robert Oppenheimer’s “tribute to the mind’s high splendor” remark.
  • People: Thomas J. Watson Jr. (IBM President); Nathaniel Rochester (chief architect); Jerrier Haddad (managing engineer); Ralph Palmer (Poughkeepsie electronics director); Cuthbert Hurd (Applied Science); James Birkenstock (Future Demands).
  • Breakthrough: IBM’s first commercial scientific computer. Watson Jr. expected 5 orders. He got 18, despite raising the price from $8,000 to $15,000 per month between the paper-plan tour and the formal announcement. (The famous “world market for maybe five computers” misquote attributed to Watson Sr. is actually Watson Jr.’s 1953 stockholders’-meeting remark, garbled.)
  • Consequence: Only 19 units ever built. Williams-tube memory (mean time between memory faults ~15 minutes); 33-instruction set; no hardware floating point; no index registers; programmers had to write self-modifying code to index arrays. The first JNWPU operational forecast, 6 May 1955, was issued on a 701. It was disappointing – the three-level QG model was too crude, the QG assumptions broke down at the equator, and the Williams tubes failed mid-run. Three years of bad forecasts, 1955-1958. But IBM’s revenue trajectory was set: from a punch-card company worth $200 million per year in 1950, IBM controlled 85% of the U.S. computer market by 1957. The 701 built IBM, even though the 701 itself was a memory-fragile, programming-hostile machine.
  • Punch: IBM’s first commercial scientific computer had a memory that crashed every fifteen minutes and a programming model that required writing programs which modified themselves while running.
  • Source: Wikipedia - IBM 701; existing Post 14 (The Machine That Built IBM) covers this in detail.

Section 4: The Stretch Catastrophe (1956-1961)

6 January 1956: Los Alamos RFP Initiates Project Stretch

  • Date: 6 January 1956 (LASL request for proposal); 5 September 1956 (first joint IBM-LASL meeting); November 1956 (formal $4.3 million contract signed for delivery by May 1960).
  • People: Stephen W. Dunwell (project director); Werner Buchholz (architecture); Frederick P. Brooks (architecture, including the eight-bit byte decision); Gene Amdahl (briefly involved before splitting to System/360 work); John Backus (consulting on programming).
  • Project: IBM committed to building a machine 100 times faster than the IBM 704 (~4 MIPS), funded primarily by the AEC for nuclear weapons design. Original budget: $4.3 million. The Stretch was meant to be the supercomputer that would consolidate IBM’s lead.
  • Source: Wikipedia - IBM 7030 Stretch; IBM Stretch project history at Mark Smotherman’s site, Clemson; Computer History Museum - IBM Stretch event

April-May 1961: Stretch Delivered to Los Alamos and Public Withdrawal

  • Date: April 1961 (Los Alamos delivery, accepted May 1961); November 1961 (Lawrence Livermore delivery); February 1962 (NSA Harvest variant 7950).
  • The failure: The machine ran at only 1.2 MIPS sustained – approximately 30 times the IBM 704, not the promised 100 times. The performance shortfall was a factor of three. February 1961 testing showed the 7030 was running user programs at 32-40 times the 704 (about half the expected speed). Watson Jr. had been getting regular updates and was, by Murray’s account, “very upset by what he perceived to be deception.”
  • Watson’s public humiliation: In May 1961, one month after the Los Alamos delivery, Watson Jr. announced at the Western Joint Computer Conference a price cut from $13.5 million to $7.78 million per machine and the immediate withdrawal of the 7030 from further sales beyond the eight customers with whom IBM was already in negotiation. Only 9 customer machines were ever delivered: Los Alamos, Lawrence Livermore, NSA (Harvest variant), Atomic Weapons Research Establishment Aldermaston, US Weather Bureau, Argonne, MITRE, IBM Yorktown, and one other. IBM lost approximately twice the original budget on Stretch.
  • The Dunwell purgatory: Stephen Dunwell, the project director, was made the public scapegoat. He was, in Murray’s phrase, “banished to a research lab.” For five years he was in IBM’s professional doghouse. Then, after System/360 (1964) became a runaway commercial success and was found to draw heavily on Stretch architectural innovations – the 8-bit byte, instruction look-ahead, pipelining, error checking – IBM rehabilitated him. At an awards banquet in April 1966, Watson Jr. publicly apologized for being unfair to Dunwell and named him an IBM Fellow. Dunwell took early retirement shortly afterward. He died 21 March 1994 in Poughkeepsie at age 80.
  • Consequence: Stretch was the most expensive single-project commercial computer failure of the era. But its architectural innovations – 8-bit addressable bytes, instruction pipelining, hardware error correction – went straight into System/360 and became the backbone of all commercial computing for 40 years. The lesson IBM took: never again pre-announce performance targets you cannot independently verify. The lesson Watson Jr. failed to take: phantom announcements are dangerous (see CDC’s 1968 antitrust suit, below).
  • Punch: IBM lost twice the budget building a computer that was a third as fast as promised, fired the project lead in disgrace, then five years later had Watson Jr. apologize publicly to him because the System/360 had been built on his work.
  • Source: Wikipedia - IBM 7030 Stretch; Dvorak News Blog - Whatever Happened to the IBM Stretch; Computer Pioneers - Stephen W. Dunwell; Wikipedia - Stephen W. Dunwell.

Section 5: The UNIVAC LARC (1955-1969) – The Other Failure

June 1960: LARC Delivered to Lawrence Livermore

  • Date: Requirements published 1955 by Edward Teller; first delivery June 1960 (Lawrence Livermore); second delivery to Navy’s David Taylor Model Basin (~1961).
  • Machines/People: UNIVAC LARC (Livermore Atomic Research Computer); Sperry Rand was the contractor.
  • The failure: Only two machines were ever built (despite an architecture supporting multiprocessing, both shipped with single CPUs). The LARC was built using surface-barrier transistors, which were already obsolete by the time the first system was delivered. Performance: addition time 4 microseconds, ~250 KIPS, 60-bit decimal words using bi-quinary coded decimal. Weight ~115,000 pounds; core memory 20,000 to 97,500 words.
  • Consequence: LARC was briefly the world’s fastest computer (1960-1961); it lost the title to the IBM 7030 Stretch in 1962. Decommissioned at Livermore December 1968; Navy’s unit turned off April 1969. The LARC was the supercomputer Edward Teller had personally specified for nuclear weapons design, and it was a commercial dead end. The lesson: by the late 1950s, a company could not build a successful supercomputer using transistor technology that was already obsolete. Sperry Rand’s UNIVAC division would never recover its early lead in scientific computing.
  • Punch: Teller’s pet supercomputer was obsolete on the day it shipped, and only two were ever built.
  • Source: Wikipedia - UNIVAC LARC.

Section 6: The Discrete-Transistor Revolution (1957-1962)

April 1955 / December 1957: IBM 608 Calculator – First All-Transistor Commercial Machine

  • Date: Announced April 1955; shipped December 1957; withdrawn April 1959.
  • People: IBM Engineering Research Labs at Poughkeepsie.
  • Breakthrough: The IBM 608 was “the world’s first all-transistorized calculator to be manufactured for the commercial market.” Over 3000 germanium transistors. 50% size reduction and 90% power reduction vs. vacuum tube models. It was a calculator, not a computer, but it was technically the first all-transistor machine IBM shipped.
  • Failure: “Only a few dozen were ever delivered.” Obsoleted within 18 months by IBM’s own faster transistor offerings (the 7070 series). It was a transitional machine – IBM was still figuring out whether to commit to transistors or stay with hybrid tube/transistor designs.
  • Consequence: The 608 demonstrated that all-transistor commercial machines could be built. IBM’s full commitment to transistors followed in the 1959 7090 (the first transistor mainframe). But the 608 itself was a small commercial failure that nobody remembers.
  • Punch: IBM’s first all-transistor commercial product, the 608, was withdrawn from sale 16 months after shipping.
  • Source: Wikipedia - IBM 608

November 1957 / Late 1958: Philco TRANSAC S-2000 – The Forgotten First All-Transistor Mainframe

  • Date: Announced 1957; first ship after autumn 1958.
  • People: Philco Corporation (Philadelphia); the company had pioneered surface-barrier transistors, which it had supplied to MIT’s TX-0 (1956) and to early DEC PDP machines.
  • Breakthrough: The Philco TRANSAC S-2000 was the first commercially produced large-scale all-transistor scientific computer (the IBM 608 was a calculator). It was a 48-bit machine, similar in architecture to the CDC 1604 that would follow. Philco had a real first-mover advantage.
  • Failure: By 1964, only 18 Model 210, 18 Model 211, and 7 Model 212 systems had been sold (43 units total) across all three S-2000 variants. Philco’s company-level failure: it could not match IBM’s deployment of resources for development, customer support, and marketing. Once IBM committed to transistors with the 7090 in late 1959, Philco was outsold catastrophically. A company that had pioneered the technology was crushed by the company that productized it.
  • Consequence: Philco was acquired by Ford Motor Company in 1961 and effectively exited large-scale computing by the mid-1960s. The first commercial transistor mainframe vendor disappeared from the market within five years.
  • Punch: The first all-transistor mainframe in the world sold 43 units before its maker was acquired by Ford.
  • Source: Wikipedia - Philco computers; Computer History Museum - Transistorized Computers Emerge

January 1960: CDC 1604 Serial Number 001 to Naval Postgraduate School

  • Date: 16 October 1959 (announced at Western Joint Computer Conference); January 1960 (serial #1 installed Spanagel Hall, Room 101A, NPS Monterey).
  • People: Seymour Cray (architect); William Norris (CEO); Frank Mullaney (founder); James Thornton; Captain Paul M. Wolff (NPS, first customer).
  • Breakthrough: The CDC 1604 was the first general-purpose digital computer ever built using transistors throughout, designed for scientific computation. 48-bit word; hardware floating point; 32,768-word core memory; ~100,000 single-address operations per second. Cost: under $1 million, against ~$2.9 million for the comparable IBM 7090. Power draw 5.5 kW vs. ~50-60 kW for the tube-era IBM 709. CDC undercut IBM on price by more than 2:1 and matched it on speed.
  • Consequence: 50 units sold by 1964. August 1960, eight months after delivery, Wolff’s team produced “the first surface weather map produced by a computer” at the Fleet Numerical Weather Facility on the same Monterey 1604. By December 1961, FNWF was distributing daily forecasts to over 100 U.S. Navy ships and shore stations across the Pacific. By 1968, a CDC 1604A was operating at the Joint Institute for Nuclear Research in Dubna (Soviet bloc). The 1604 was the machine that made operational fleet weather forecasting possible, and it was the financial bridge that funded CDC’s larger 6600 supercomputer.
  • Punch: The first serial-number-001 of the first all-transistor scientific computer in the world produced the first computer-generated weather map eight months later, at half the price of IBM’s competing machine.
  • Source: Wikipedia - CDC 1604; existing Post 29 (Serial Number One) is the canonical narrative.

30 November 1959 / 1962: IBM 7090 and the IBM 7094

  • Date: First IBM 7090 installation December 1959; transistorized successor to IBM 709 (1958 vacuum tube). IBM 7094 announced 1962.
  • Specs: ~50,000 germanium alloy-junction transistors. 36-bit word. ~100 KFLOPS (7090); ~200 KFLOPS (7094). Six times faster than the 709. 1960 system cost: $2.9 million, monthly rental $63,500. Half the rental of the vacuum-tube 709.
  • Significance: The IBM 7090 became the workhorse of NMC at Suitland from approximately 1960 onward. The three-layer hemispheric model – the first model that produced operational 3D forecasts in the United States – became operational in 1962 on the 7090 (per NCEP/EMC history). The 7090 was used at NASA for Mercury, Gemini, and Apollo space tracking. In October 1962, during the Cuban Missile Crisis, the IBM 7090s and 709s of the time were the primary computational substrate for U.S. military meteorology, missile-effects calculations, and operational space tracking. The Defense Atomic Support Agency had a CDC 1604 running during the crisis for nuclear-effects work; the 7090s at NMC were producing operational weather forecasts that fed into Strategic Air Command planning. For the first time in history, computer-generated forecasts mattered at the highest level of geopolitical decision-making.
  • Consequence: The 7090 and 7094 were withdrawn from sale 14 July 1969 but remained in productive use for over a decade afterward. The IBM 7090 family was the most commercially successful transistor-era scientific mainframe.
  • Source: Wikipedia - IBM 7090; NCEP/EMC History; existing Post 14 covers the JNWPU/NMC machine sequence.

Section 7: The CDC 6600 / IBM 360 Confrontation (1963-1965)

22 August 1963: CDC 6600 Announcement

  • Date: 22 August 1963 (CDC press announcement); 28 August 1963 (Watson Jr.’s “thirty-four people including the janitor” memo); September 1964 (serial #1 to Lawrence Livermore); 14 January 1965 (CERN, first European supercomputer); December 1965 (NCAR serial #7).
  • People: Seymour Cray (architect); James Thornton (logic designer, scoreboard architect); Les Davis (mechanical/packaging); Dean Roush (refrigeration); Thomas J. Watson Jr. (IBM CEO).
  • Breakthrough: 60-bit word; 1-microsecond core memory cycle; 100 ns CPU clock; 3 MFLOPS peak; ten parallel functional units dispatched by Thornton’s hardware Scoreboard – the first commercial implementation of out-of-order instruction issue with hardware-managed dependency tracking. Ten Peripheral Processor Units in a “barrel” architecture for the OS, the first hardware-multithreaded design. Freon-cooled cylindrical plus-sign cabinet. At any given instant, the 6600 was doing more work in parallel than any commercial machine before it.
  • The Watson memo: Six days after CDC’s announcement, Watson Jr. dictated the now-canonical memo: “I understand that in the laboratory developing this system there are only thirty-four people ‘including the janitor.’ Of these, fourteen are engineers and four are programmers, and only one person has a Ph.D., a relatively junior programmer. Contrasting this modest effort with our vast development activities, I fail to understand why we have lost our industry leadership position.” Cray’s reported reply, when the memo was passed back to him: “It seems like Mr. Watson has answered his own question.”
  • Consequence: ~100 6600s sold between 1964 and 1972 at $7-10 million apiece. Five-year reign as the world’s fastest computer. The 6600 was the supercomputer of choice at Livermore, Los Alamos, Berkeley, NYU Courant, CERN, NCAR, NASA Ames, NSA, and dozens of universities. The IBM antitrust nightmare started here: Watson Jr.’s memo and the subsequent Jenny Lake retreat in September 1963 produced two strategic decisions – commit to System/360, and launch a “6600 killer” project that would eventually become the IBM 360/91 / 360/92 phantom announcement.
  • Punch: A thirty-four-person team in rural Wisconsin, funded by a startup whose total founding capital was $1.2 million, beat the world’s largest computer company to the world’s most powerful computer.
  • Source: Wikipedia - CDC 6600; existing Post 30 (Thirty-Four People) is canonical.

7 April 1964: IBM System/360 – The Most Consequential Computing Announcement of the Era

  • Date: 7 April 1964 (announcement event in Poughkeepsie, attended by reporters from over 100,000 customers/journalists/technologists in 165 U.S. cities).
  • People: Thomas J. Watson Jr. (CEO); Gene Amdahl (chief architect); Frederick P. Brooks (project lead for software, OS/360); Bob O. Evans (system planning).
  • Breakthrough: The most consequential commercial computing announcement of the era. Six computer models announced simultaneously (Models 30, 40, 50, 60, 62, 70), with 44 peripherals and a unified instruction set across the entire family. First time a manufacturer offered binary-compatible computers spanning an order of magnitude in performance.
  • The bet: System/360 cost IBM an estimated $5 billion in development. Watson Jr. famously called it the “five-billion-dollar gamble” and a “bet-the-business” move. Orders exceeded forecasts: more than 1,000 systems were purchased in the first month after announcement.
  • The OS/360 disaster: Frederick Brooks managed software development. The OS/360 project was, in Brooks’s later memoir The Mythical Man-Month (1975), the founding example of how software engineering schedules cannot be accelerated by adding programmers. Watson Jr. asked Brooks “why it was so much harder to manage software projects than hardware projects” – a question that produced the most influential book in software engineering. The hardware succeeded; the software was years late and the disaster of the era.
  • Consequence: System/360 dominated commercial mainframe computing for 40 years. System/360-compatible hardware (in the form of the IBM Z-series) is still being shipped in 2026. The 8-bit byte that Brooks had inherited from Stretch became the universal unit of computer memory. The architecture of every subsequent IBM mainframe is direct descent from the 7 April 1964 announcement.
  • Punch: The most consequential commercial computer announcement in history was a $5 billion bet on six binary-compatible models, accompanied by an OS that was years late and 1000 customer orders in the first month.
  • Source: Wikipedia - IBM System/360; Computer History Museum - April 7 IBM Announcement; IEEE Spectrum - Building System/360

Section 8: The IBM 360/91 / 360/92 Phantom Announcement and CDC’s Antitrust Suit

November 1964: IBM Announces the 360/92 (and Then Quietly Renames It the 360/91)

  • Date: April 1964 (Model 90 was a “footnote” in the original System/360 announcement); November 1964 (Model 92 announcement); 1965-1966 (Model 92 quietly renamed Model 91).
  • People: Gene Amdahl (architect, then split off); John Cocke; Robert Tomasulo (instruction-issue logic); Jack Bertram (Project Y/ACS lead); Lynn Conway; Herb Schorr.
  • The phantom announcement: IBM, panicked by the CDC 6600, announced the Model 92 (and the broader Project X / Project Y / ACS programs) in November 1964 with spectacular performance specifications. The Model 92 was scaled back to the Model 91, a more modest machine. A succession of high-end machines – 67 (1966), 85 (1969), 91 (1967, “anticipated as the 92”), 95 (1968), 195 (1971) – replaced the phantom 92.
  • The phantom: an even better example. Project Y / Advanced Computer Systems was an internal IBM “6600 killer” running 1965-1969, originally targeting 10x the 6600. It cost approximately $15 million in software development alone (~$145 million in 2025 dollars). Cancelled May 1969 because Gene Amdahl concluded “there was no way its sales could turn a profit” and because Amdahl’s competing AEC/360 design promised better cost-effectiveness. IBM never shipped its 6600 killer.
  • The 360/91: 15 ever produced, 4 of those for IBM internal use. Tomasulo’s algorithm for register renaming and out-of-order execution was the lasting legacy – it would lie dormant for 28 years before being revived in 1995 by Intel’s Pentium Pro and becoming the universal organizing principle of every modern CPU.
  • Consequence: A “full page of the CDC suit, entitled ‘Paper Machines and Phantom Computers,’ is devoted to this charge regarding IBM’s practice of announcing machines with specifications it could not and never intended to meet.” When CDC decided to enter the scientific computer market, IBM “announced a computer for the same market with spectacular specifications, but they were not met, with at least four delays in the delivery date and less than a dozen ever delivered.”
  • Punch: IBM announced the supercomputer of the decade, never built it, used the announcement to dampen CDC’s sales, and got sued for it.
  • Source: Wikipedia - IBM Advanced Computer Systems project; Wikipedia - IBM System/360 Model 91; Wikipedia - IBM System/360; CHM Revolution - CDC 6600’s Five Year Reign

December 1968: CDC Files Antitrust Suit Against IBM

  • Date: December 1968 (filing); January 1973 (settlement).
  • People: William Norris (CDC CEO); Tom Barr (Cravath Swaine & Moore, IBM defense); David Boies and Jay Gerber (CS&M associates); Nicholas Katzenbach (IBM General Counsel from 1969).
  • The suit: CDC accused IBM of monopolizing the computer market in violation of Sherman Act Section 2. The principal allegation: IBM’s “phantom” announcement of the 360/91 (and its predecessors and successors) in November 1964 had adversely affected CDC’s sales of the 6600 even though IBM was slow to actually deliver the announced machine. The suit included a “Paper Machines and Phantom Computers” section dedicated to documenting IBM’s practice of pre-announcing products it could not (or did not intend to) deliver.
  • The settlement: In January 1973, IBM settled with CDC for cash and assets totaling over $600 million. The settlement included transfer of an IBM computer service company at less than market value. The settlement was approximately one hundred times the size of CDC’s original founding capital ($1.2 million in 1957).
  • Consequence: The CDC settlement funded CDC’s diversification through the 1970s and 1980s, including the PLATO computer-based-education programme that Norris championed personally. It established that pre-announcement of products that don’t ship can be antitrust violation. It was the largest computer-industry antitrust settlement of its era and one of the loudest signals that IBM’s monopolistic behaviour had limits. (The Department of Justice’s separate antitrust suit against IBM, filed January 1969, would drag on for 13 years and end in dismissal in 1982.)
  • Punch: The thirty-four-person Wisconsin laboratory’s parent company sued IBM for phantom announcements and won $600 million.
  • Source: dwkcommentaries - The IBM Antitrust Litigation; CHM Revolution - CDC 6600’s Five Year Reign.

Section 9: Mid-1960s Misfires (1965-1968)

16 August 1965: IBM System/360 Model 67 / TSS Time-Sharing Disaster

  • Date: Announced 16 August 1965 via IBM “blue letters”; first shipped May 1966; TSS/360 cancelled in 1968, reprieved in 1969, cancelled again in 1971; Model 67 hardware withdrawn 15 March 1977.
  • People: Driving demand: MIT (Project MAC); University of Michigan (where MTS was developed); General Motors; Bell Labs; Princeton; Carnegie Tech; Naval Postgraduate School. Researchers: Bruce Arden, Bernard Galler, Frank Westervelt, Tom O’Brien (foundational paper January 1966).
  • The promise: IBM announced TSS/360 simultaneously with the Model 67, describing it as a “powerful operating system” offering “virtually instantaneous access” via time-sharing. The Model 67 was IBM’s response to MIT’s Multics project and the broader academic interest in interactive time-sharing.
  • The flop: TSS/360 was repeatedly delayed. Cancelled 1968. Reprieved 1969. Cancelled definitively 1971. “IBM’s failure to deliver TSS/360 as promised opened the door for others to develop operating systems.” The Model 67 hardware shipped, but the time-sharing OS that was supposed to make it useful was never delivered in production form.
  • Consequence: IBM’s failure created a vacuum into which alternatives stepped. MTS (Michigan Terminal System), CP/CMS, and VP/CSS were all developed by Model 67 customers as substitutes for TSS. CP/CMS evolved into VM/370, IBM’s eventually successful virtualization OS. The Model 67 spawned a thriving time-sharing community that IBM had initially underestimated and that ran on alternatives to IBM’s own promised software. Time-sharing as an industry was created by IBM customers fixing IBM’s promise. (The longer-term consequence: TSS’s failure and the success of the customer-developed alternatives are part of what convinced Watson Jr. to fund the unbundling of IBM’s software in 1969 – the strategic decision that created the commercial software industry.)
  • Punch: IBM promised time-sharing in 1965, cancelled it in 1968, reprieved it in 1969, cancelled it again in 1971, and the customers who had bought the hardware built better operating systems themselves.
  • Source: Wikipedia - IBM System/360 Model 67

1966 / 1968: ILLIAC III – The Academic SIMD Failure

  • Date: Built 1966 at the University of Illinois; destroyed in a fire in 1968 (“a Variac shorting on one of the wooden-top benches”); rebuilt during the early 1970s; ultimately abandoned.
  • People: Bruce H. McCormick (project lead throughout); John P. Hayes (input-output channel control unit logic).
  • The project: A fine-grained SIMD pattern recognition computer with a specialized “Pattern Articulation Unit” as its core parallel-processing component. Built initially for image processing of bubble chamber experiments to detect nuclear particles, later extended to biological imaging. The team also explored multiple-radix arithmetic concepts.
  • The fire and the failure: 1968 fire destroyed the original build. After reconstruction in the early 1970s, “the core parallel-processing element of the machine, the Pattern Articulation Unit, was successfully implemented” – but the project ultimately failed and was abandoned. The computer never produced sustained scientific output.
  • Consequence: ILLIAC III is the canonical example of an academic computer project of the 1960s that aimed too high, was destroyed by accident, was rebuilt at great cost, and produced no lasting research output. (Slotnick’s ILLIAC IV – a separate project at the same university – would have its own well-documented difficulties through 1972-1981. Both ILLIAC III and ILLIAC IV are important not for what they produced but for what they failed to produce: the SIMD philosophy did not commercially work in the late 1960s and 1970s, and would not return until the GPU era of the 2010s.)
  • Punch: A Variac shorting on a wooden bench burned down the world’s most ambitious academic SIMD computer in 1968.
  • Source: Wikipedia - ILLIAC III

June 1969: CDC 6800 Renamed CDC 7600 – The Compatibility Break

  • Date: 6800 design largely complete 1966-67; renamed 7600 in late 1966 or early 1967 when Cray decided to break compatibility; announced 1967; first delivery June 1969; ~75 units sold.
  • People: Seymour Cray (architect); James Thornton (still at CDC).
  • The breakthrough: 36 MFLOPS peak (12x the 6600). Deep instruction pipelining; small fast core memory + large slow core memory hierarchy (the conceptual ancestor of modern cache). Freon cooling.
  • The break: The 7600 had originally been named the CDC 6800, but during design, “the designers determined that maintaining complete compatibility with the existing 6000-series machines would limit how much performance improvement they could attain and decided to sacrifice compatibility for performance.” Cray broke compatibility with the entire 6000 family, requiring all 6600 customers to recompile and validate their software for the 7600. The CDC 6800 never shipped under that name.
  • The reliability disaster: The 7600 introduced severe reliability problems. “Both LLNL and NCAR reported that the machine would break down at least once a day, and often four or five times.” Heat density was beyond existing cooling technology. “Acceptance at installation sites took years while the bugs were worked out, and while the machine generally sold well enough given its ‘high end’ niche, it is unlikely the machine generated any sort of real profits for CDC.”
  • Consequence: The 7600 became the workhorse scientific computer of the integrated-circuit era nonetheless. NCAR ran serial #12 from May 1971 to April 1983 – almost twelve years. The compatibility break is the single most important architectural lesson of the era: even the best architect (Cray) breaks compatibility when the technology demands it. The customers absorb the cost.
  • Punch: The CDC 6800 was renamed the CDC 7600 mid-design when Cray decided performance trumped compatibility, and customers paid for it with two years of acceptance-test bugs and a machine that crashed five times a day.
  • Source: Wikipedia - CDC 7600

Section 10: Concept of “Computer-Generated Forecasts Mattered Geopolitically” (1962)

October 1962: Cuban Missile Crisis and Computer-Generated Forecasts

  • Context: For the present post, the most useful framing is that the IBM 7090 era at NMC Suitland (~1960-1965) was the first time computer-generated forecasts had geopolitical consequence. Specific evidence for the missile crisis itself:
    • The Defense Atomic Support Agency (DASA) at the Pentagon used a CDC 1604 during the Cuban Missile Crisis of October 1962 for nuclear-effects calculations and contingency-targeting work. The DASA 1604’s specific role during the crisis is partially declassified; the basic fact that the machine was running and that Cray’s CDC team was on standby for emergency support is in the public record.
    • The IBM 7090s at NMC Suitland were running the operational three-layer hemispheric model from 1962 onward, providing forecasts to U.S. military aviation and Strategic Air Command planners during the 13 days of the crisis (16-28 October 1962).
    • Telstar 1 launched 10 July 1962 – IBM 1401 computers handled the satellite test message of 25 October 1962 between Endicott NY and La Gaude France. Computer-mediated international communication was, by October 1962, an operational reality.
  • Punch for the post: The hardware of the discrete-transistor era – IBM 7090, CDC 1604 – was the operational computational substrate during the closest the Cold War came to nuclear war. The forecasts, calculations, and tracking that ran on those machines fed into the highest-level decisions of the crisis.
  • Source: existing Post 29 (Serial Number One); Wikipedia - Telstar; NCEP/EMC History – the “three-layer hemispheric model 1962” attribution.
  • Caveat: Direct primary-source documentation of operational meteorological computing during the Cuban Missile Crisis is limited because much of the relevant material remains classified or is buried in DoD archives. The strongest claim that can be safely made: by October 1962, the U.S. military and atmospheric sciences had transistorized scientific computing as operational infrastructure, and that operational reality was new in 1962. Three years earlier, NMC was running on the IBM 704; the Navy on an ERA 1101. By 1962 the discrete-transistor era was the substrate of operational meteorology.

Section 11: Watson Sr.’s 1952 Defense Calculator Quote (corrects the user’s prompt)

29 April 1952: Watson Sr. Announces the Defense Calculator at IBM Annual Stockholders Meeting

  • Date: 29 April 1952. (NB: the user’s prompt referenced “the 1954 Watson Sr. press conference quote about the IBM 701” – the canonical primary statement is actually the 29 April 1952 stockholders meeting quote, which is what the user likely means.)
  • The quote (Watson Sr.): “Our progress in electronics convinced us one year ago that we had in our company the ability to create for the Defense Department, and the defense industries, a computer of advanced design which could be of major service to our national defense effort. We began planning and building such a machine, which we believe will be the most advanced, most flexible high-speed computer in the world.”
  • The context: This was Watson Sr.’s formal acknowledgement that IBM had pivoted to electronic computing. It came at a time when Watson Sr. (born 1874) was 78 years old and had publicly resisted the electronic-computer transition for years; Watson Jr. (born 1914) was the actual driver. The 1952 stockholders meeting announcement was, in effect, the moment when the old IBM acknowledged the new IBM.
  • Watson Jr.’s 1953 follow-up at the next year’s stockholders meeting: “As a result of our trip, on which we expected to get orders for five machines, we came home with orders for 18.” This is the line that, as Wikipedia documents and as Post 14 covers in detail, was later misattributed to Watson Sr. as “I think there is a world market for maybe five computers” – a quote Watson Sr. never said.
  • Source: IBM history - Thomas J. Watson Sr.; historyofcomputercommunications.info - The Entrance of IBM 1952; Wikipedia - IBM 701

Section 12: Magnetic Core Memory Cost Curve and Displacement (1953-1973)

Magnetic Core Memory: The Cost Curve That Made It Universal

  • Period of dominance: 1953-1976 (approximately).
  • Cost curve:
    • 1953: tested but not-yet-strung cores cost $0.33 each.
    • 1970: price per core fell to $0.0003 (three hundredths of a cent) – a 1000x reduction in 17 years.
    • Overall trajectory: from ~$1.00 per bit in early production to ~$0.01 per bit by 1970.
  • Peak production: By 1970, IBM was producing 20 billion cores per year.
  • The displacement: Intel 1103 DRAM (1970, 1024 bits in 18-pin DIP, p-MOS three-transistor cell) launched at $60 per chip. Production improvements drove the price below $4 per chip by 1973 – less than 1% of the equivalent core cost. By the end of 1971 the 1103 was the best-selling semiconductor in the world; within two years, 14 of the 18 major computer manufacturers, including HP, DEC, Honeywell, and CDC, had adopted it.
  • Consequence: By the late 1970s essentially every new computer was shipping with semiconductor RAM. Magnetic core lingered for another decade in two specialized markets: military and aerospace systems (radiation-hardness, non-volatility) and a handful of legacy minicomputer lines. The terminology survives: the Unix command for writing a process’s memory to disk is core dump; the resulting file is a core file.
  • Punch: The memory technology that powered every weather model from Whirlwind (1953) through the IBM 7090 (1959) through the CDC 6600 (1964) through the CDC 7600 (1969) was supplanted in three years by a single Intel chip released in October 1970.
  • Source: Wikipedia - Magnetic-core memory; existing Post 28 (Memory in Every Forecast) is canonical.

Section 13: The Transition Frame (1958-1968)

What the Discrete-Transistor Era Made Possible (1958-1968) – the Era 2 Story Arc

  • The transition: Three orders of magnitude smaller than vacuum tubes; three orders less power; three orders less heat; two orders more reliable. The CDC 1604 (1960) at 5.5 kW; the IBM 709 (its tube-era predecessor) at 50-60 kW. Power and heat dropped by an order of magnitude on the same scientific computing performance.
  • What it allowed:
    • Operational fleet weather forecasting at FNWF (August 1960 onward).
    • Operational three-layer hemispheric weather forecasting at NMC (1962 onward).
    • Operational five-day global forecasts (1965 onward).
    • First true atmospheric general circulation models – Kasahara-Washington 1967 NCAR paper, on the CDC 6600.
    • GFDL coupled-model experiments on UNIVAC 1108 (1967-1973).
    • The IBM System/360 commercial transition that built modern computing.
  • What it could not do:
    • Ensemble forecasting at scale (no spare compute for repeating runs with perturbed initial conditions).
    • Quantification of the Lorenz two-week predictability limit (Lorenz had identified it on his LGP-30 in 1961, but operationalizing it would wait for the Cray-1 era).
    • Time-sharing at scale – the IBM 360/67 / TSS disaster proved that the era’s hardware was not yet ready for interactive multi-user operation.

Failure Modes of the Era (Cluster Summary)

  1. Memory technology that did not scale: Williams tubes (15-min MTBF), mercury delay lines (sequential access), magnetic drums (slow). All collapsed by 1955-1960 against magnetic core.
  2. Vacuum tube reliability ceiling: Even the best vacuum-tube machine (AN/FSQ-7 SAGE) was technologically obsolete on the day of its 1958 commissioning. The 49,000-tube count was a ceiling, not a base.
  3. Pre-announcement that didn’t ship: IBM’s 360/91 / 360/92 phantom announcements forced the CDC antitrust suit and a $600 million settlement in 1973. IBM’s ACS project never shipped its 6600 killer despite spending $15 million on software.
  4. Performance promises that didn’t land: IBM Stretch (3x slower than promised); Watson Jr.’s 1961 public withdrawal; Dunwell’s five-year purgatory.
  5. Compatibility breaks that customers absorbed: CDC 6800 → 7600 (1967); UNIVAC abandoning its early lead through Sperry-Rand corporate dysfunction.
  6. Academic SIMD failures: ILLIAC III (1968 fire, never-recovered); ILLIAC IV (separate project, would not deliver useful science until ~1976).
  7. Time-sharing at scale: TSS/360 announcement 1965, cancelled 1968, reprieved 1969, cancelled 1971.
  8. Niche supercomputer commercial dead ends: UNIVAC LARC (only 2 built, obsolete on delivery); Philco TRANSAC S-2000 (43 units, parent acquired by Ford).

Successes of the Era (Cluster Summary)

  1. 8 August 1953: Magnetic core memory at Whirlwind. The single most important architectural step of the period.
  2. 21 May 1952 / 7 April 1953: IBM 701 announcement and Oppenheimer dedication. The 701 built IBM.
  3. 6 May 1955 / 1958: First operational JNWPU forecast on IBM 701 / IBM 704; routine operational NWP established by 1958.
  4. 26 June 1958: First SAGE direction center operational at McGuire AFB; SAGE pays for IBM’s manufacturing infrastructure.
  5. December 1959 / January 1960: First IBM 7090 / first CDC 1604 (serial #1 to NPS Monterey); transistor scientific computing arrives.
  6. August 1960: First computer-generated weather map at FNWF Monterey.
  7. 1962: Three-layer hemispheric operational model on IBM 7090 at NMC; computer-generated forecasts become geopolitically consequential during Cuban Missile Crisis.
  8. 22 August 1963 / September 1964 / December 1965: CDC 6600 announcement; Livermore serial #1; NCAR serial #7. The first true supercomputer.
  9. 7 April 1964: IBM System/360 announcement. The most consequential commercial computing announcement of the era.
  10. 1967: Kasahara-Washington NCAR Community Climate Model on the CDC 6600 – the first true atmospheric general circulation model on a real supercomputer.

Quick-Reference Index of Punchy Moments (for Drafting Use)

Date Event Punch
29 Aug 1949 Soviet RDS-1 detonation Saves Whirlwind from cancellation
13 Jun 1949 Forrester’s coincident-current notebook Architecture of core memory
8 Aug 1953 Core memory at Whirlwind MTBF: 4 hours/day → 2 hours/week
21 May 1952 IBM 701 announced Watson Jr. expected 5 orders, got 18
6 May 1955 First JNWPU operational forecast “Very disappointing and not usable”
6 Jan 1956 Stretch contract initiated $4.3M for 100x the IBM 704
Apr 1957 First FORTRAN compiler shipped 25,000 lines of machine code
Dec 1957 IBM 608 ships First all-transistor commercial machine
Late 1958 Philco TRANSAC S-2000 ships First all-transistor mainframe – 43 units total
Dec 1959 IBM 7090 first installation $2.9M, 6× faster than 709
Jan 1960 CDC 1604 serial #1 to NPS Monterey Half the price of IBM 7090
Jun 1960 LARC delivered to Livermore Obsolete on delivery; 2 ever built
Aug 1960 First computer weather map at FNWF Cycle-to-cycle automation
Apr 1961 Stretch delivered to Los Alamos 1.2 MIPS sustained, not 4
May 1961 Watson Jr. cuts Stretch price to $7.78M, withdraws $13.5M → $7.78M; only 9 ever shipped
1962 Three-layer hemispheric model on IBM 7090 at NMC First operational 3D forecast
Oct 1962 Cuban Missile Crisis DASA CDC 1604 + NMC IBM 7090 operational
22 Aug 1963 CDC 6600 announced “Thirty-four people including the janitor”
28 Aug 1963 Watson Jr.’s memo Cray’s reply: “Watson has answered his own question”
7 Apr 1964 IBM System/360 announced $5B bet; 1000+ orders in first month
Sep 1964 CDC 6600 serial #1 to Livermore $7-10M, 3 MFLOPS
Nov 1964 IBM 360/92 announced (phantom) Renamed 360/91; never built as 92
Dec 1965 CDC 6600 serial #7 to NCAR Ran until May 1977
16 Aug 1965 IBM 360/67 + TSS announced TSS cancelled 1968, reprieved 1969, cancelled 1971
Apr 1966 Watson Jr. apologizes to Dunwell IBM Fellow honor; 5 years after disgrace
1967 Kasahara-Washington NCAR GCM published First GCM on a real supercomputer
1967 CDC 6800 renamed 7600 Cray breaks compatibility
1968 ILLIAC III destroyed by fire A Variac shorting on a wooden bench
Jun 1969 CDC 7600 first delivery Crashes 5× per day for 2 years
May 1969 IBM ACS cancelled $15M software effort, never shipped 6600 killer
Dec 1968 CDC files antitrust suit against IBM $600M settlement in January 1973

Sources Index (consolidated for citation use in the post)

Wikipedia primary sources (all accessed 2026-05-10):

Computer History Museum and ETHW:

IBM corporate sources:

Antitrust and litigation:

NWP institutional sources:

Books to cite:

  • Charles J. Murray, The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer (Wiley, 1997)
  • Emerson W. Pugh, Memories That Shaped an Industry: Decisions Leading to IBM System/360 (MIT Press, 1984)
  • C. J. Bashe, L. R. Johnson, J. H. Palmer, E. W. Pugh, IBM’s Early Computers (MIT Press, 1986)
  • T. J. Watson Jr. with Peter Petre, Father, Son & Co. (Bantam, 1990)
  • Frederick P. Brooks, The Mythical Man-Month (Addison-Wesley, 1975, anniversary edition 1995)
  • James E. Thornton, Design of a Computer: The Control Data 6600 (Scott Foresman, 1970)
  • Paul N. Edwards, A Vast Machine: Computer Models, Climate Data, and the Politics of Global Warming (MIT Press, 2010)
  • George Dyson, Turing’s Cathedral: The Origins of the Digital Universe (Vintage, 2012)

Notes for the Author (Drafting Guidance)

  1. The Soviet bomb date: The user’s prompt has “1958-08-29” which is likely a typo for 1949-08-29. The Soviet RDS-1 test in August 1949 is what triggered the air defense panic that saved Whirlwind from cancellation in 1950 and led to SAGE’s 1958 commissioning. Either correct the date in the post or check whether a 1958 event was meant (Sputnik was 4 October 1957; the U-2 incident was 1 May 1960; nothing significant on 29 August 1958).

  2. Cuban Missile Crisis weather computing: The strongest defensible claim is that the 1962 IBM 7090 era at NMC was the first time computer-generated forecasts had geopolitical consequence. Direct primary-source documentation of operational meteorology during the 13 days of the crisis is sparse; the post should frame this carefully. The DASA 1604 (a CDC machine, not an NMC weather computer) is the strongest documented Cuban-Missile-Crisis computer use.

  3. Telstar / weather connection: There is no direct connection between Telstar and weather forecasting in 1962. Telstar was a communications satellite. The IBM 7090 at NMC was producing weather forecasts on a parallel timeline, but the two are not the same story. The user’s prompt grouped them together; the post should treat them as parallel developments, not joint.

  4. The Watson Sr. 1954 quote: The user’s prompt referenced “the 1954 Watson Sr. press conference quote about the IBM 701” – the canonical primary statement is actually the 29 April 1952 stockholders meeting quote. There is no specific 1954 quote that fits the description. The post should use the 1952 quote (or Watson Jr.’s 1953 stockholders’ meeting “five, eighteen” quote) and clarify the date.

  5. The most punchy single moments to feature:

    • Stretch’s collapse (Watson Jr. publicly cuts price by 42%, withdraws product, fires Dunwell, then apologizes to him 5 years later) – this is the era’s most dramatic single failure arc and deserves 200-300 words.
    • The 360/92 phantom and CDC’s $600M antitrust settlement – the era’s most consequential failure, and not currently in the post. Worth 200-300 words.
    • The CDC 6800 → 7600 rename – a small technical decision that produced two years of customer pain and is the canonical case study in compatibility-vs-performance tradeoffs.
    • TSS/360 cancellation cycle – IBM cancelled the same OS three times. Customers built better alternatives (MTS, CP/CMS) themselves.
    • ILLIAC III’s 1968 Variac fire – a single colorful sentence that captures the academic-supercomputer disaster of the era.
    • The 49,000 vacuum tubes of SAGE – verified, with the punchline that SAGE’s $8-12 billion cost was 4-6× the Manhattan Project.
    • 8 August 1953 core memory at Whirlwind – maintenance time dropped 50-100x; the single architectural step that made all subsequent atmospheric science possible.