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.

CDC 1604 – Research File

Summary in one paragraph

The Control Data Corporation Model 1604 was a 48-bit, fully-transistorised, single-address scientific computer designed primarily by Seymour Cray. CDC announced it on 16 October 1959, and the first machine – Model #1, Serial #1 – was delivered in January 1960 to the U.S. Naval Postgraduate School in Monterey, California, where Cray himself supervised the installation. It is widely cited as the first commercially successful, fully-transistorised, large-scale scientific computer. About fifty units were sold by the mid-1960s. It was the launch product of CDC, the machine that established Cray’s reputation as a computer architect, and the platform on which the Fleet Numerical Weather Facility produced what was widely advertised as “the first surface weather map produced by a computer.” NCAR’s first computer, however, was NOT the 1604; it was the CDC 3600, delivered in November 1963.

Origin (1957–1960)

  • CDC was founded in September 1957 by William Norris and a team of former employees of the scientific-computing division of Sperry Rand (itself created when Remington Rand bought ERA in 1952 and Sperry then bought Remington Rand in 1955). [Wikipedia: Control Data Corporation; Murray, The Supermen, 1997]
  • Norris became CEO. Cray, the lead architect on the ERA 1103 / UNIVAC 1103, joined CDC slightly later: he stayed at Univac long enough to finish detailed design of the AN/USQ-17 (the U.S. Navy “Unit computer,” NTDS predecessor), then walked across to CDC. [Wikipedia: Naval Tactical Data System; Murray]
  • CDC’s first office was at 501 Park Avenue, Minneapolis. The legend that the model number “1604” was 501 + 1103 (Cray’s previous design) is repeated everywhere – but a 1975 oral history with CDC engineers recorded by the Charles Babbage Institute reports that insiders “laughed and responded: ‘It was quite popular at the time that this was the origin… We’ve never been able to substantiate it.’” CDC’s official explanation: the original design goal was to support 16K of memory and 4 tape units. [Wikipedia: CDC 1604, citing CBI OH 321]
  • In 1958, Cray built a 6-bit, 64-word prototype called “Little Character” to validate the modular packaging and transistor-circuit-card approach he wanted for CDC’s first machine. The Computer History Museum holds it. The results were “encouraging” enough that CDC committed to the 48-bit production system that became the 1604. [CHM, “The Father of Supercomputing”; CHM Revolution exhibit on Little Character]
  • CDC was at this point uncertain about its product direction – the CHM notes the company “even explored electronic cash registers” before Cray’s prototype redirected it toward serious scientific machines.
  • The first commercial 1604 contract: a U.S. Navy Bureau of Ships order in June 1959, worth approximately $2.5 million for the initial systems. [Grokipedia, citing CDC corporate history]
  • Public announcement: 16 October 1959. [Westport Tech Museum; Grokipedia]
  • First delivery: January 1960, to the Naval Postgraduate School (NPS), Monterey, California, room 101A of Spanagel Hall. Customer Acceptance Test was passed on 16 January 1960. [BRL61 survey; CHM “Weather by Computer”]

Architecture

Word, registers, and instruction format

  • Word length: 48 bits. [Wikipedia; BRL61 survey; CDC Reference Manual]
  • Two instructions per word, packed into 24 bits each. Instruction format: 6-3-15
    • 6 bits operation code
    • 3 bits index designator (selects index register B1–B6 for memory access; or condition for branches; designator 7 means indirect addressing)
    • 15 bits memory address (or shift count)
  • Number of operation codes: 62 (BRL61 lists “62 instructions decoded”; Grokipedia confirms “62 basic instructions”; opcodes 00 and 77 are reserved/fault). The successor CDC 3600 added five op codes for full upward compatibility. [BRL61; Wikipedia; Grokipedia]
  • Registers:
    • A – 48-bit accumulator
    • Q – 48-bit auxiliary arithmetic register (used as low half of double-length AQ for multiply/divide; holds mask in logical operations)
    • P – 15-bit program counter
    • B1–B6 – six 15-bit index registers
    • Plus internal storage-address registers S1/S2 for the two memory banks, restoration registers Z1/Z2, an exchange register R, and an external-function register
  • Indirect addressing: built in. [BRL61]

Memory

  • Magnetic-core memory. Maximum: 32 768 words of 48 bits = 192 kilobytes (the 32 767 figure on Wikipedia’s infobox is a rounding/labelling artifact). [Wikipedia; BRL61]
  • BRL61 lists base-system options at 8 192, 16 384, or 32 768 words.
  • Cycle time per bank: 6.4 microseconds. Two interleaved 16K-word banks (odd/even addresses) phased 3.2 microseconds apart, giving an effective average access time of 4.8 microseconds. [BRL61; Wikipedia; Grokipedia]
  • 1 500 000 magnetic cores total. [BRL61]

Arithmetic

  • One’s complement for fixed-point integers; range ±(2^47 – 1).
  • Floating-point: hardware native, format 1-11-36 (1 sign, 11 biased exponent, 36 significand). [Wikipedia; BRL61]
  • Add: 4.8–9.6 microseconds (parallel arithmetic, including memory access). [BRL61]
  • Multiply: 25.2 + 0.8N microseconds (where N is the number of ones in the multiplier). [BRL61]
  • Divide: 63.6–66.4 microseconds. [BRL61]
  • Operation: synchronous, parallel, concurrent (arithmetic could overlap with control). [BRL61]

Performance number cited to customers

  • About 100 000 operations per second (~0.1 MIPS) is the figure most often quoted in CDC marketing and contemporary BRL surveys. [Wikipedia; CHM “Weather by Computer”; BRL61 (4.8 microsecond effective cycle on representative programs)]
  • An advertised “208 kHz” CPU clock figure refers to the unswapped memory-access frequency, not effective instruction rate.
  • One cute architectural feature: the top three bits of the accumulator drove a digital-to-analog converter wired to a tube audio amplifier in the operator’s console. Programmers could compose music or use the tone as a debugger – an endless repetitive musical phrase meant the program had hung in a loop. [Wikipedia]

I/O architecture

  • Six independent buffer channels (three input, three output) plus a high-speed transfer channel. [BRL61]
  • Program-interrupt facility for peripheral signalling and overflow detection. [BRL61]
  • The CDC 160 (12-bit, released 1960; standalone version: CDC 160-A) commonly served as I/O processor / front-end computer to the 1604. The 160-A is sometimes called “arguably the first minicomputer.” [Wikipedia: CDC 1604, CDC 160 series]

Logic Technology

  • Fully transistorised. Diode-transistor logic. The BRL61 entry tallies for an entire 1604 system: 100 000 diodes, 25 000 transistors, 1 500 000 magnetic cores. [BRL61]
  • Germanium transistors. CDC selected germanium for its switching speed and low power dissipation in 1958–1959. Cray would later report that he could not push germanium parts much faster, and the 6600 in 1964 moved to silicon transistors. [Murray, The Supermen; Computer History Museum: “1961: Silicon Transistor Exceeds Germanium Speed”]
  • Switching speed of the system clock: about 200 kHz (core memory cycle dominates timing). [Grokipedia; BRL61]
  • The “37-cent transistor” anecdote (low confidence). Several secondary sources (history-of-computing summaries citing Murray’s The Supermen) report that the 1604 prototype was built with 37-cent transistors purchased over the counter from “the Radio Shack of its day,” and that Cray bought so many that General Transistor doubled the price. The detail is plausible (Cray was known for cost-engineering and General Transistor was a real germanium-transistor supplier of the period), but the citation chain back to a primary source is weak. The Murray book is the canonical source – I have not been able to verify the page directly. Flag as anecdote, attribute carefully.
  • Cordwood module construction. The 1604 used a stackable, modular plug-in chassis design. The famous tightly-packed “cordwood” modules with ~600 000 transistors per machine are associated more with the CDC 6600 (1964), but the 1604 already used the modular printed-circuit-card approach Cray validated with Little Character. The 1604’s stackable, removable circuit-card system established a CDC tradition. [Grokipedia; Thornton, Design of a Computer: The Control Data 6600, 1970]
  • Power: 5.5 kW at 208 V, 60 Hz for the central system; entire installation including air conditioning was substantially higher. [Wikipedia; BRL61]
  • Heat removal: forced-air, with site-prep requirements typically including a 5-ton air-conditioner unit at the manufacturer’s spec, larger in heavier configurations. The transistor approach made this trivial compared to the 1103/IBM 704 generation. [BRL61]
  • Heat / power comparison vs. vacuum-tube generation: vacuum-tube scientific machines of comparable role (IBM 709 etc.) drew more like 50 kW; the 1604 cut that by an order of magnitude. [Grokipedia]
  • Reliability: Grokipedia (citing CDC) notes “zero transistor failures” in early production deployment in 1960 – a startling change from the every-few-hours tube failures of the 704/709/1103 era. [Grokipedia, citing internal CDC reports]

Comparison to Predecessors and Contemporaries

Predecessors (Cray’s own and the field)

  • ERA 1101 (1950, Atlas I as built for the Navy): Cray’s first design environment at ERA. Vacuum-tube, drum-memory, ~350 instructions/second on representative work. [Wikipedia: UNIVAC 1101]
  • ERA 1103 / UNIVAC 1103 (October 1953): the machine for which Cray was first credited with major design work. About 3 900 vacuum tubes plus 9 000 diodes, 36-bit words, electrostatic Williams-tube primary memory of 1024 words plus a 16384-word magnetic-drum store, clock speed 500 kHz. [Wikipedia: UNIVAC 1103]
  • UNIVAC 1103A (Athena variant, 1956): added core memory, replacing Williams tubes, and was the version used at the Air Force Cambridge Research Center.

1959–1960 contemporaries

  • IBM 704 (1954): vacuum-tube, 36-bit, ~40 thousand instructions per second on FORTRAN code – a generation behind the 1604.
  • IBM 7090 (delivered November 1959): IBM’s transistorised replacement for the 704/709. 36-bit word, transistorised, ~100 thousand floating-point operations per second on the Gibson mix benchmark; price ~$2.9 million per system, lease ~$63 500 per month. The 7090 was the 1604’s main commercial competitor; it shipped about two months ahead of the 1604’s first delivery, but the 1604’s 48-bit word length gave it an edge on double-precision floating-point work. [Wikipedia: IBM 7090; Grokipedia]
  • Philco TRANSAC S-2000 (1958, “Model 210”): the first transistorised scientific computer offered as a product by a manufacturer. Surface-barrier-transistor logic. 48-bit word, fully transistorised, ~2000 lb. By 1964: 18 Model 210s, 18 Model 211s, 7 Model 212s sold. The 1604 was somewhat slower than the most-advanced Philco models on paper but matched or exceeded them in practice and had a far longer commercial life. [Wikipedia: Philco computers]
  • UNIVAC LARC (1960, delivered to LLNL and Navy): ambitious dual-processor specialist machine, ~$6 million per system, 350 kVA power draw, 3 000 sq ft footprint. Compared to the 1604’s two cabinets and ~1000 kg footprint, the LARC was a different class – and a commercial flop, with only two ever built. [Grokipedia]
  • Ferranti Atlas (1962, Manchester): transistorised, 48-bit, 128 index registers, 1.6 microsecond float-add, ~700 thousand instructions/second peak; pioneered virtual memory (“one-level store”). Atlas was technically ahead of the 1604 but was a one-off research project, not a commercial product line; only three were ever built. The 1604’s distinctiveness in 1960 was that it shipped as a real product – what made it stand out was that Cray, almost alone, had translated the transistor revolution into a system anyone could buy and run in under 100 kW. [Grokipedia; Wikipedia: Atlas]

Performance

  • Customer-cited “100 000 operations per second.” This is the round figure that appears in CDC marketing material, BRL surveys, and contemporary press coverage and that customers used in their specs. [Wikipedia; CHM “Weather by Computer”]
  • Per-operation breakdown (BRL 1961, including memory access):
    • Add (parallel, fixed-point): 4.8–9.6 microseconds
    • Multiply: 25.2 + 0.8N microseconds (data-dependent)
    • Divide: 63.6–66.4 microseconds
  • Floating-point hardware native, included in the basic system at no extra cost (this was rare in 1960; many contemporaries charged for floating-point as an option).
  • Comparison to IBM 7090: Grokipedia cites a Gibson-mix benchmark of approximately 139 KIPS for the 7090 vs. ~81 KIPS for the 1604. Flag as low confidence: I have not been able to verify the original Gibson-mix source; raw benchmarks from this era depended hugely on configuration and compiler. The widely-told story is that the 1604 was “competitive with” or “outperformed” the 7090 on certain double-precision floating-point work because the 48-bit word avoided the need for multi-precision arithmetic, while the 7090 was faster on simpler integer-heavy work. [Grokipedia; Murray]

Cost and Sales

Pricing (manufacturer list, BRL 1961 survey)

Memory configuration Purchase price 1-year lease/month 3-year lease/month
8 192 words $750 000 $22 500 $18 750
16 384 words $830 000 $25 000 $20 750
32 768 words $990 000 $30 100 $24 750
  • Wikipedia’s headline price of $1 030 000 is for the fully-loaded 32 768-word (“192 kilobytes”) configuration as listed in the 1964 Computer Survey – close to but slightly above the 1961 list price. [Wikipedia, citing BRL 1964 survey]
  • Add-on prices: Model 1607 magnetic-tape subsystem (4 drives, 30 KC) was $145 000; the Model 1606 high-speed printer was $110 000; the Model 1605 IBM-compatibility adaptor was $70 000. [BRL61]
  • The actual NPS configuration cost approximately $800 000 (with 4 tape drives, photoelectric reader, console, teletype punch). [BRL61]
  • The headline figure to quote in writing: “around one million dollars per machine, depending on configuration” – for the 32 768-word system that was the standard buy. The 1961 listed price was $990 000; later versions (CDC 1604-A) crossed the $1 million line. [BRL61; Wikipedia 1964 survey citation]
  • Comparison to IBM 7090: ~$2.9 million purchase, $63 500/month lease in 1960. The 1604 was substantially cheaper. [Wikipedia: IBM 7090]

Sales volume

  • About 50 systems by 1964, the most commonly cited figure. [Wikipedia; Grokipedia; Britannica; multiple secondary sources]
  • BRL 1961 (early) listed: 6 produced, 6 in operation, 10 in production, 6 on order, 1 per month anticipated production rate, 9 months delivery time.
  • Grokipedia’s wording is “approximately 50 by 1965”; Wikipedia says “50+ units” sold; Westport Tech Museum says “approximately 50 units.” All three figures are consistent.
  • For the post angle: “about fifty” is the safe, accurate figure.

Major Deployments and Customer List

U.S. Navy installations (the dominant customer)

  • U.S. Naval Postgraduate School, Monterey, California – Spanagel Hall, Room 101A, January 1960. Model #1, Serial #1. Personally installed and certified by Cray. Used both for instruction and for the FNWF weather work that took place on the same campus. A second 1604 was installed in 1961 in the converted lobby on the first floor. [CHM “Weather by Computer”; CDC brochure 1963; NPS history pages]
  • Fleet Numerical Weather Facility (FNWF, Monterey) acquired its own 1604 in 1961 (see weather section below). [CDC brochure 1963]
  • Fleet Operations Control Centers in Hawaii (Pearl Harbor / Kunia), London, and Norfolk, Virginia for major weather-prediction and fleet-coordination workloads. [Wikipedia; DBpedia]
  • FOCCPAC at Kunia, Oahu, Hawaii: by 1969 was running an “Automated Control Environment” (ACE) cluster of five CDC 160As supervising four CDC 1604s in a multi-tasking network – one of the largest 1604 installations. [Wikipedia]

Strategic / classified

  • Pentagon, Defense Atomic Support Agency (DASA): one 1604 used during the Cuban Missile Crisis (October 1962) to predict possible Soviet strike scenarios against the United States. [Wikipedia; multiple secondary sources]
  • National Security Agency (NSA, Fort Meade): NSA acquired multiple 1604s starting in September 1960 (“CDC-1604(1) with a special WELCHER attachment”), with subsequent units in February 1961, March 1962, January 1963, and July 1963. Cryptanalytic use. Also used at NSA on the BOGART project, which had begun the transition to solid-state computing. [Grokipedia, citing NSA declassified history pamphlet on Seymour Cray]
  • U.S. Naval Ordnance Laboratory, Corona, California: 1604 delivered December 1960, missile-trajectory analysis. [Grokipedia]
  • Minuteman I (LGM-30A/B) ICBM ground stations: certain single-silo designs used two redundant CDC 1604s for pre-launch guidance computation, including current weather-driven targeting adjustments. Block models exist at the Octave Chanute Aerospace Museum, Rantoul, Illinois. [Wikipedia]
  • Lawrence Livermore National Laboratory acquired a CDC 1604 in 1962 for nuclear-weapons simulation – a secondary system; LLNL’s primary machines were UNIVAC LARC and IBM 7030 Stretch, but the 1604 served alongside. [Grokipedia, citing LLNL history]

Universities and research

  • National Bureau of Standards, Boulder, Colorado – used for radio-propagation, radio-standards, and cryogenics research. [BRL61]
  • University of Illinois, Coordinated Science Laboratory, Urbana: a 1604, donated by William Norris, became the host of PLATO III (third-generation computer-based-instruction system, mid-1960s, supporting up to 20 student terminals simultaneously). [Wikipedia: PLATO; CSL archives; Grokipedia]
  • University of Minnesota – FORTRAN classes for general computing instruction. [Grokipedia]
  • The 1964 BRL survey lists installations at the U.S. Air Force, Navy, Army Signal Corps and various U.S. universities; the full census is dispersed across multiple BRL editions. [BRL64]

Commercial

  • Marathon Oil Company, Findlay, Ohio (1960): the Masquerade text-mining application, one of the earliest. [Wikipedia]
  • University Computing Corporation, Dallas (1963): a service-bureau 1604 ran seismic-data processing for Sun Oil and other oil companies. [Grokipedia]

Note on CERN

  • The CDC 1604 was NOT delivered to CERN. CERN’s first major CDC machine was a CDC 6600 in January 1965, followed by a CDC 6500 and CDC 3800. The 3000-series successors did go to several European labs, but I have not been able to find a primary-source confirmation of any 1604 export to CERN. The post-and-some-secondary-sources occasional claim “CERN had a 1604” appears to confuse the 1604 with the later 3000-series machines. [CERN Timeline; CERN Document Server records]

The Soviet Sale (Dubna)

This needs careful unpacking. There are at least three distinct stories in circulation:

  1. The fact (well-attested): “The 1604 design was used by the Soviet nuclear weapons laboratory. The BESM-6 computer, which entered production in 1968, was designed to be somewhat software compatible with the CDC 1604, but it ran 10 times faster and had additional registers.” [Wikipedia: CDC 1604, citing Impagliazzo & Proydakov, Perspectives on Soviet and Russian Computing, Springer, 2011]
  2. The acquisition story (variously reported): a CDC-1604 reached the USSR in 1968. Russian-language sources connected to the BESM-6 emulator project state that “The USSR bought the CDC-1604 in 1968” and that the developers of the BESM-6 FORTRAN translator were tasked with making it compatible with CDC-1604 programs. [Russian Virtual Computer Museum; Topwar.ru “BESM saga”; Hackaday on the BESM-6 emulator]
  3. The Wikipedia talk-page rumour (low confidence): an editor on Talk:CDC 1604 alleged “the Soviets stole a CDC 1604 from West Germany and used it at their nuclear research lab.” This is uncited and likely apocryphal; it does not appear in any of the published references I checked.

The Dubna / JINR claim specifically: the Joint Institute for Nuclear Research at Dubna, USSR, did receive Western computers under controlled export – by the late 1960s and into the 1970s the JINR computer centre included an SDS-1604A. Important: SDS-1604A was made by Scientific Data Systems, NOT the same as the CDC 1604. SDS used the model number 1604A as one of its Sigma-series predecessors, so any reference to “an SDS-1604A at Dubna” is not the same machine. This is a documented confusion in the literature. [Wikipedia: Joint Institute for Nuclear Research; Wikipedia: SDS Sigma series]

Bottom line for the post:

  • The cleanest verifiable claim is: a CDC 1604 reached the Soviet Union in or by 1968, was used at a Soviet nuclear-research facility, and inspired the architecture and FORTRAN compatibility of the BESM-6.
  • The 1965-1966 / Commerce Department export-license / Dubna-specific story is low confidence: I could not locate a primary-source citation for that exact narrative. The user’s research brief’s hypothesis is plausible but I cannot confirm it from the sources I checked.
  • I would NOT name “Dubna” with confidence in the post unless I find better sourcing. “A Soviet nuclear-research facility” or “a Soviet research institute” is the safer wording.

Atmospheric Science and NWP Use

This is the most important section for the blog post; I want to be very careful.

NCAR (CRITICAL FACT-CHECK)

  • NCAR’s first computer was NOT a CDC 1604. NCAR’s first computer was a CDC 3600, delivered in November 1963 and installed in a University of Colorado building under construction at 3215 Marine Street, Boulder. By early 1964 NCAR’s Computing Facility was running production on the 3600. The 3600 had 32 700 48-bit words of magnetic core, supported a FORTRAN compiler, and ran roughly 1 MIPS. [NCAR/CISL: “NCAR Supercomputing History – 3600”; Cray-History.net “HPC at NCAR Past Present and Future”]
  • NCAR was incorporated in 1960 but its computing facility predates only by paper plans; the first machine arrived in late 1963. NCAR did some early computing on time leased on outside CDC machines (Boulder NBS, U. Minnesota etc.) but did not own a 1604. DO NOT claim the 1604 was NCAR’s first computer; it was not.
  • The successor lineage at NCAR was: CDC 3600 (1963) -> CDC 6600 (mid-1960s as second machine) -> CDC 7600 (serial #12 in May 1971) -> Cray-1 (serial #3 in July 1977).

Fleet Numerical Weather Facility (THE actual NWP story for the 1604)

This is the strongest weather-NWP story for the post.

  • March 1959: the U.S. Navy assigned the “Numerical Weather Problems Group” to Monterey under Captain Paul M. Wolff.
  • January 1960: CDC 1604 Model #1, Serial #1 installed at NPS, Spanagel Hall 101A. Cray personally certified it.
  • February 1961: the group was renamed the Fleet Numerical Weather Facility (FNWF). A second 1604 went to FNWF directly (not just the NPS instructional unit).
  • August 1960 (between those two dates): the Monterey Peninsula Herald reported that FNWF (then still NANWEP) had produced “the first surface weather map to be produced by a computer,” cutting hemispheric forecast compilation “from hours to minutes” and claiming it was “40 percent more accurate than old hand methods.” [CHM “Weather by Computer”; Medium/CHM Core “Weather Prediction Goes Digital”]
  • By December 1961: FNWF was routinely transmitting computer-produced weather and oceanographic forecasts to over one hundred Navy units worldwide by radioteletype and facsimile. [CDC brochure “Weather by Computer,” 1963]
  • Forecast products generated on the 1604:
    • hemispheric analyses every 12 hours
    • sea-surface temperature analyses and prognoses
    • wind, wave, and swell forecasts
    • combined sea analyses
    • surface-current analyses
    • ocean-atmosphere heat-exchange analyses
    • mixed-layer-depth predictions (used for submarine-detection sonar tactics)
    • hurricane steering and trajectory prediction
  • Computational scale (CDC’s 1963 brochure): “one complete analysis and forecast cycle involves over two billion individual computations.” Each hemispheric analysis covered 4 000 geographic grid points. Forecasts extended 48 hours forward at hourly resolution.
  • A specific example given in CDC’s brochure: “A storm forecast of such magnitude uses 8000 instructions, six minutes computing time for a 48-hour forecast.”
  • Commander Paul Wolff was the first officer-in-charge of FNWF and is a named, quotable character in the story. Cmdr. Ted Hesse was the hurricane-forecasting officer.
  • The 1604 stayed in service well past the IBM 360/67 acquisition in 1967: the original Model #1, Serial #1 was transferred to FNWF and moved to Point Pinos, Pacific Grove, where it continued in service for archival weather-data storage. [CHM “Weather by Computer”]
  • FNWF moved off NPS in 1974, was renamed Fleet Numerical Meteorology and Oceanography Center (FNMOC) in 1993, and continues to operate today. The CDC 1604 was the founding machine of what is now FNMOC. This is the strong NWP angle for the post.

JNWPU and operational NWP

  • The Joint Numerical Weather Prediction Unit (JNWPU) at Suitland, Maryland – the U.S. civilian/military operational NWP centre established in 1954 – ran on IBM 701 (1955), IBM 704 (~1958), and IBM 7090 (~1960-onwards), and never on a CDC 1604. So the operational NWP centre at Suitland is NOT in the 1604 customer list.
  • The 1604 thus belongs in the NWP story specifically as the Navy/Fleet computer at Monterey, NOT as a civilian-NWP/Weather-Bureau machine. Suitland was an IBM shop in this period.

Anecdotes

  • “Little Character,” 1958. Cray’s 6-bit, 64-word prototype to prove the modular-card transistor logic approach. Now in the Computer History Museum collection. The system worked well enough that CDC committed to a full 48-bit production design. [CHM]
  • The 37-cent transistor (low-confidence anecdote, attribute carefully). Murray’s The Supermen reportedly recounts that Cray purchased the prototype’s transistors at 37 cents each from an electronics store, and that General Transistor doubled the price after he bought so many. Verify before quoting; I have not seen the page.
  • Cray installing Serial #1 himself. Edward Norton Ward, the first computer technician at NPS, recalled Cray “like a master pianist, ran through the test programs” during installation in Spanagel Hall in January 1960. [CHM “Weather by Computer”]
  • The Chippewa Falls Laboratory, 1962. In 1962, Cray demanded a remote laboratory away from CDC’s Minneapolis bureaucracy. Norris debated, then backed him; CDC built him a laboratory on land Cray owned in his hometown of Chippewa Falls, Wisconsin, about 100 miles east of Minneapolis on a river bluff. From this point onwards, all subsequent Cray-led machines (CDC 6600, 7600, 8600, and Cray-1, Cray-2 at Cray Research) were designed at Chippewa Falls, not Minneapolis. This anecdote belongs at the END of the 1604 story – the 1604 was designed in Minneapolis, but its commercial success was the leverage Cray used to demand his own lab. [Murray; HandWiki: Seymour Cray]
  • The acoustic accumulator. The 1604 console included a tube audio amplifier connected through a D/A converter to the top three bits of the accumulator. Programmers used it for music; debuggers used it to detect infinite loops. [Wikipedia]
  • James E. Thornton’s role. Thornton joined ERA in 1950 (B.S., U. Minnesota), left with other engineers to form CDC in 1958, and was a core 1604 engineer alongside Frank Mullaney, George Henson, and others. Thornton’s most famous work came later (chief architect of the CDC 6600, with the 1970 book Design of a Computer: The Control Data 6600), but he and Cray collaborated on the 1604. The exact division of labour between Cray-as-architect and Thornton-as-implementer on the 1604 is less crisp than the iconic Cray/Thornton split on the 6600. [Wikipedia: James E. Thornton; Murray]
  • The “501 Park Avenue + 1103 = 1604” legend. Repeated everywhere; the CBI oral history notes it was popular but never substantiated. CDC’s official explanation: 16K of memory + 4 tape units. [Wikipedia, citing CBI OH 321]

Successors

  • CDC 924 (1961, designer Cray): a 24-bit cut-down version sharing peripheral compatibility with the 1604. Roughly $180 000 per system. About 12 sold by 1964. Notable because the first 924 was delivered to NASA. [Wikipedia]
  • CDC 160 / 160-A (1960, 1962): 12-bit minicomputer used as I/O front-end for 1604s and as a small standalone scientific computer. The 160-A is sometimes called “the first minicomputer.” Designed by Cray over an extended weekend, according to legend. [Wikipedia]
  • CDC 3000 series (announced May 1962, first delivery June 1963):
    • Lower 3000 series: 24-bit (3100, 3200, 3300, 3500) – broader commercial market.
    • Upper 3000 series: 48-bit (3400, 3600, 3800) – direct upgrade path for 1604 customers, “largely compatible” with the 1604 plus five extra op codes.
    • CDC 3600: first delivered June 1963, ~1 MIPS, 32K x 48-bit core. NCAR’s first computer (November 1963). Cray did the basic architecture; he handed the 3000 series off to other engineers (notably Thornton) to complete while he started on the 6600. [Wikipedia: CDC 3000 series]
  • CDC 6600 (delivered late 1964): the breakthrough machine – the world’s first true supercomputer. Silicon transistors, 60-bit word, ~3 MFLOPS, 10 functional units in parallel. Designed at Chippewa Falls by a 34-engineer team under Cray, with Thornton as principal collaborator. About 100 sold. Famously, IBM’s Tom Watson Jr. wrote his 1965 internal memo asking how a 34-person team in rural Wisconsin had outperformed IBM’s massive engineering organisation. This is the post that follows the 1604 post. [Wikipedia: CDC 6600; Murray]

The clean lineage to quote in the blog post: 1604 (1960, transistors prove themselves) -> 3600 (1963, full upgrade path, NCAR’s first machine) -> 6600 (1964, the breakthrough) – with the 924 and 160-A as side-branches and the 3000 series as the bridge. Cray was already moving on to the 6600 when the 3000 series shipped.

Sources

Primary

  • Control Data 1604 Computer, Vol. 1: Description and Operation, December 1960. Bitsavers PDF: http://bitsavers.trailing-edge.com/pdf/cdc/1604/031a_1604_Computer_Vol_1_Description_and_Operation_Dec60.pdf
  • Control Data 1604-A Computer Reference Manual, 245a rev. 5/63, Control Data Corporation, Minneapolis, 1963. Bitsavers PDF: http://bitsavers.org/pdf/cdc/1604/245a_1604A_RefMan_May63.pdf
  • Characteristics of the Model 1604 Computer, Control Data Corporation, May 1961. Internet Archive (bitsavers): https://archive.org/details/bitsavers_cdc1604018heModel1604Computer196105_4859064
  • Weather by Computer, Control Data Corporation, sales brochure, 1963. Computer History Museum brochure collection. PDF: http://s3data.computerhistory.org/brochures/cdc.weatherbycomputer.1963.102641261.pdf
  • Survey of Domestic Electronic Digital Computing Systems, Ballistic Research Laboratory Report 1115, U.S. Army, 1961. Online transcription including the CDC 1604 entry: https://www.ed-thelen.org/comp-hist/BRL61-c.html
  • A Survey of Domestic Electronic Digital Computing Systems, BRL Report 1234, U.S. Army, 1964 (lists CDC 1604A): http://ed-thelen.org/comp-hist/BRL64-c.html#CONTROL-DATA-1604A

Secondary – canonical history

  • Charles J. Murray. The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer. New York: John Wiley & Sons, 1997. ISBN 978-0471048855. [The canonical popular history of CDC and Cray; verify the 37-cent transistor and Cerberus/General-Transistor anecdotes against this volume’s pages.]
  • James E. Thornton. Design of a Computer: The Control Data 6600. Glenview, Ill.: Scott, Foresman, 1970. Computer History Museum scan: https://archive.computerhistory.org/resources/text/CDC/cdc.6600.thornton.design_of_a_computer_the_control_data_6600.1970.102630394.pdf [Has 1604 history in introduction; CDC engineering culture context.]
  • John Impagliazzo and Eduard Proydakov, eds. Perspectives on Soviet and Russian Computing. Springer, 2011. ISBN 978-3-642-22816-2. [Cited by Wikipedia for the BESM-6 / 1604 software-compatibility claim.]
  • Charles Babbage Institute Oral History 321: “Reminiscences of computer architecture and computer design at Control Data Corporation,” interview with Neil R. Lincoln and 18 CDC engineers, c. 1975. https://conservancy.umn.edu/server/api/core/bitstreams/523b280d-ea3d-4abd-ab6e-315bf32adf03/content
  • Charles Babbage Institute Control Data Corporation Collection – Historical Timeline. http://www.cbi.umn.edu/collections/cdc/histtimeline.html

Online encyclopedia and museum sources

  • “CDC 1604,” English Wikipedia. https://en.wikipedia.org/wiki/CDC_1604
  • “Control Data Corporation,” English Wikipedia. https://en.wikipedia.org/wiki/Control_Data_Corporation
  • “Seymour Cray,” English Wikipedia. https://en.wikipedia.org/wiki/Seymour_Cray
  • “James E. Thornton,” English Wikipedia. https://en.wikipedia.org/wiki/James_E._Thornton
  • “BESM-6,” English Wikipedia. https://en.wikipedia.org/wiki/BESM-6
  • “CDC 1604,” Grokipedia. https://grokipedia.com/page/CDC_1604
  • “CDC 1604 computer,” Britannica. https://www.britannica.com/topic/CDC-1604

Computer History Museum

  • “Typical CDC 1604 installation,” CHM Revolution exhibit. https://www.computerhistory.org/revolution/supercomputers/10/22/15
  • “Control Data Corporation, ‘Little Character’ prototype,” CHM Revolution. https://www.computerhistory.org/revolution/supercomputers/10/22/18
  • “The Father of Supercomputing,” CHM Revolution. https://www.computerhistory.org/revolution/story/22
  • “Weather by Computer,” CHM Blog, by David A. Laws. https://computerhistory.org/blog/weather-by-computer/
  • David A. Laws, “Weather Prediction Goes Digital. Fleet Numerical Weather Facility…,” CHM Core (Medium). https://medium.com/chmcore/weather-prediction-goes-digital-596f5425c043
  • “October 16: CDC Introduces 1604 Computer,” CHM This Day in History. https://www.computerhistory.org/tdih/october/16/

NCAR, FNMOC, and weather-history sources

  • NCAR Computational and Information Systems Lab (CISL), “CDC 3600”: https://www.cisl.ucar.edu/ncar-supercomputing-history/3600
  • NCAR/CISL, “NCAR Supercomputing History” (overview): https://www.cisl.ucar.edu/ncar-supercomputing-history
  • “HPC at NCAR: Past, Present and Future,” Cray User Group 2010 Proceedings, archived at Cray-History.net: https://cray-history.net/wp-content/uploads/2022/01/HPC-at-NCAR-Past-Present-and-Future.pdf
  • Naval Postgraduate School, “70 Years in Monterey”: https://nps.edu/70-years-in-monterey
  • Fleet Numerical Meteorology and Oceanography Center, public site: https://www.fnmoc.navy.mil/home/

NSA / classified-history sources

  • “Seymour Cray and NSA,” NSA declassified-document pamphlet, 5 October 2018. https://www.nsa.gov/portals/75/documents/news-features/declassified-documents/history-today-articles/10%202018/05OCT2018%20SEYMOUR%20CRAY%20and%20NSA.pdf [Note: returned 403 in some attempts but is publicly listed on NSA.gov.]
  • “Computer Development: Cray Supercomputers,” NSA Cryptologic Museum: https://www.nsa.gov/History/National-Cryptologic-Museum/Exhibits-Artifacts/Cryptologic-Museum/Current-Exhibits/igphoto/2002138193/

Soviet-context sources

  • “BESM-6,” Russian Virtual Computer Museum: https://www.computer-museum.ru/english/besm6.htm
  • “Joint Institute for Nuclear Research,” Wikipedia and JINR official site: https://en.wikipedia.org/wiki/Joint_Institute_for_Nuclear_Research ; https://www.jinr.ru/main-en/

Westport / Generation-of-Computer summaries

  • Westport Tech Museum, “Today in Computer History (October 16, 1959): The CDC 1604 is Released”: https://www.westporttechmuseum.com/blogs/techtastic/today-in-computer-history-october-16-1959-the-cdc-1604-is-released

Confidence flags

HIGH confidence (multiple primary or near-primary sources agree):

  • 48-bit word, 32 768 word maximum core, 6.4 microsecond cycle, 4.8 microsecond effective access, 62 instructions, 6-3-15 instruction format, six 15-bit index registers, hardware floating-point, indirect addressing, ones’ complement
  • Add 4.8–9.6 microseconds, “100 000 operations per second” customer-facing figure
  • ~50 systems sold by 1964; first delivery January 1960 to NPS Monterey; CDC founded September 1957; public announcement 16 October 1959
  • Pricing in 1961: $750 000–$990 000 depending on memory; ~$1 030 000 for full 32K-word system in 1964 survey
  • 25 000 transistors and 100 000 diodes per system, fully germanium-transistor logic
  • FNWF / Fleet Numerical Weather Facility used the 1604 for hemispheric weather forecasting starting 1960; first surface weather map produced by a computer (August 1960); routine forecasts to 100+ Navy units by December 1961
  • NCAR’s first computer was the CDC 3600 in November 1963, NOT the 1604
  • BESM-6 (1968) was designed for software compatibility with the 1604
  • DASA Pentagon used a 1604 during the Cuban Missile Crisis
  • Chippewa Falls Laboratory founded 1962 at Cray’s demand
  • Successor lineage: 924 (1961), 3000 series (1963), 6600 (1964)

MEDIUM confidence (strong secondary sourcing but I want to double-check before quoting verbatim):

  • The “37-cent transistors from the Radio Shack of its day” / “Cray bought so many General Transistor doubled the price” anecdote – attributed to Murray’s The Supermen but I have not verified the exact page
  • The “501 Park Avenue + 1103 = 1604” legend (CBI oral history says the engineers themselves never substantiated it)
  • LLNL acquired a 1604 in 1962 – one secondary source, not in BRL61
  • The Gibson-mix benchmark figures (139 KIPS for 7090 vs. 81 KIPS for 1604) – specific numbers from one secondary source
  • “Zero transistor failures” in 1960 production – attributed to internal CDC reports

LOW confidence (treat as anecdote or open question):

  • The exact 1965-1966 Dubna sale story with Commerce Department export licence: I could not find a primary-source citation. The cleanest claim I can defend is “a CDC 1604 reached the Soviet Union by 1968 and influenced BESM-6 architecture and software compatibility.”
  • “Soviets stole a 1604 from West Germany” – Wikipedia talk-page rumour, no published source
  • The exact NSA acquisition dates from September 1960 onwards (September 1960, February 1961, March 1962, January 1963, July 1963) – listed by Grokipedia citing the NSA’s published history but the NSA PDF I tried to fetch returned 403; I would want to retrieve and quote it directly
  • That CERN ever had a 1604 (probably did not – CERN’s first CDC was the 6600 in 1965)
  • Whether any 1604 was at the Joint Numerical Weather Prediction Unit at Suitland (almost certainly not – JNWPU was an IBM shop)