CDC at UT Austin and IBM at Exxon Houston

There's a curious parallel between UT Austin's CDC hardware and Exxon Houston's IBM hardware during the sixties and seventies, bookended by a shared IBM era in the fifties and a shared Cray era in the eighties. CDC and Cray were members of a family of Minnesota companies (ERA, CDC, Cray) that, along with its UNIVAC relatives, was a vigorous competitor to IBM in the engineering, scientific, national lab, and cryptography fields. In a sense, UT Austin moved from IBM to the ERA tradition in the sixties, and Exxon Houston followed in the eighties. UT Austin's early move was due to David Young's being firmly in the ERA tradition from his work at Ramo-Wooldridge (TRW) in the fifties. He arrived at UT in 1958 and led the acquisition of the CDC 1604 in 1960 and CDC 6600 in 1966. The CDC Cyber hardware that UT acquired in the seventies was essentially updated versions of the CDC 6600, based on the same 60-bit architecture and running similar code.

There were multiple important connections between computing at UT Austin, Exxon, Rice, and Houston. A sign of these connections was the story of how, in 1958, Humble Oil in Houston (now Exxon) donated its IBM Card-Programmed Electronic Calculator to UT Austin. UT’s Al Matsen was a consultant for Exxon Houston and New Jersey for over thirty-five years. The CPC was a landmark gift and a direct result of Matsen’s extensive ties. To bypass bureaucratic paperwork, Matsen, his graduate students, and other faculty physically carried the heavy machine components into Welch and installed it themselves. Exxon had acquired the CPC in 1952 and used it to implement ground-breaking subsurface reservoir simulations and the beginnings of the ADI Alternating Direction Implicit techniques for Finite Difference Methods. This work put Exxon, Rice University, and Houston in a leading position for subsurface modeling and computational engineering and science.

ADI was forged in late 1953 out of urgent commercial necessity by Peaceman, Rachford, and Douglas. Driven by the pragmatic need to simulate oil reservoirs for high-stakes drilling decisions, they bypassed academic idealism in favor of industrial utility. Rather than chasing elegant theorems, they engineered ADI as a brilliant, gritty algorithmic hack, splitting complex multi-dimensional problems into a sequence of cheap, one-dimensional coordinate sweeps to circumvent both the memory bottlenecks of early hardware and the finicky tuning required by SOR. The alignment of Exxon with IBM, and David Young’s association with UNIVAC and Control Data Corporation, mirrors the structural, financial, and philosophical divides of the early computing era. [1]

The IBM CPC was not a computer in the modern sense. It was a hybrid electro-mechanical system. It consisted of an IBM 402 or 417 Accounting Machine (the printer/controller) connected to an IBM 604 Electronic Calculating Punch (the arithmetic unit) and an electromechanical storage unit. It functioned as a decentralized network of specialized units rather than a unified stored-program architecture and was fundamentally incapable of holding both the data and the instructions required for the Simplex Method. Consequently, the program existed not as a digital state within the machine, but as a physical sequence of punched cards. This required the operator to function as a manual control unit, physically re-entering card decks to execute the iterative loops essential for finding an optimal solution within a linear system. 

The development of the CPC itself actually originated from clandestine, user-driven engineering at Northrop Aircraft rather than inside IBM's own research labs. In late 1946, a specialized computing group at Northrop led by engineers Greg Toben, Bill Woodbury, and Rex Rice was tackling complex aerospace calculations, such as jet propulsion and guided missile trajectories, which vastly outstripped the capacity of standard accounting machines.

Because true stored-program computers were not yet available, the Northrop team decided to build their own hardware solver by merging two leased IBM machines: the new IBM 603 Electronic Multiplier, which was fast but lacked sequencing control, and the older IBM 405 Accounting Machine, which was slow at math but had excellent card-reading and printing capabilities. In direct violation of their IBM rental agreements, the Northrop engineers took the protective covers off the machines, exposed their internal wiring, and physically linked the two units together. They affectionately dubbed their makeshift, hybrid creation the poor man's ENIAC.

This prototype, which the engineers nicknamed Betsy, was remarkably powerful but suffered from physical instabilities. The multiplier section would occasionally freeze mid-operation, trapping card decks inside. To clear the jam, the operators often resorted to physical force. In one famous incident, an engineer was told to "Kick it, Gib", and his literal kick drove a heavy metal cover directly into a 60-ampere fuse block, causing a massive, hazardous shower of sparks. When Northrop's founder, Jack Northrop, reached out to IBM's CEO Thomas Watson to demand manufacturing support and standard parts for their modified system, IBM realized the commercial potential of the hybrid concept. IBM immediately flew Woodbury and his colleague George Fenn to New York to present their 603 and 405 combo. IBM's engineers then standardized the physical interfaces, replaced the 603 with the newer 604 calculating punch, and officially announced the commercial Card-Programmed Electronic Calculator in May 1949.

Even after the commercial release, Northrop continued to drive the CPC's evolution. Early users struggled because instruction cards had to point to highly specific, hardwired microprograms on a physical plugboard, making it almost impossible to share programs. To break this hardware bottleneck, Northrop's Rex Rice engineered a general-purpose control panel. By wiring a generalized set of logical pathways and math routing systems directly into the board, programmers could write various mathematical applications entirely on standard card decks without needing to manually rewire the plugboard for every new problem.

The CPC can even be traced a few years further back, to 1943 Los Alamos and the race to build the atomic bomb. All of the pieces that would become the CPC three years later were already being brought together at Los Alamos and even earlier at Columbia University, and it’s extremely likely that some of the Northrop researchers had been present at Los Alamos and possibly Columbia. Here's a description of the arrival at Los Alamos of the pieces of what soon would become known as the CPC, and Richard Feynman’s reaction.

Feynman, frustrated, turned to Nicholas Metropolis, a mustached Greek mathematician who later became an authority on computation and numerical methods, and said, “Let’s learn about these damned things and not have to send them to Burbank.” (Feynman grew a temporary mustache, too.) They spent hours taking apart new and old machines for comparative diagnosis; learned where the jams and slippages began; and hung out a shingle advertising, “Computers Repaired.” Bethe was not amused at this waste of his theoreticians’ time. He finally ordered a halt to the tinkering. Feynman complied, knowing that within weeks the shortage of machines would change Bethe’s mind. Escalation of the computation effort came in the fall of 1943 with an order to IBM for business machines to be delivered to an unknown location: three 601 multipliers, one 402 tabulator, one reproducer-summary punch, one verifier, one keypunch, one sorter, and one collator. Astronomers at Columbia had been experimenting with punch-card computing before the war. A multiplier, an appliance the size of a restaurant stove, could process calculations in large batches. Electrical probes found the holes in the cards, and operations could be configured by plugging groups of wires into a patchboard. Among the computation-minded at Los Alamos, the prospect of such machines caused excitement. Even before they arrived, one of the theorists, Stanley Frankel, set about devising improvements: for example, tripling the output by rearranging the plugs so that three sets of three- or four-digit numbers could be multiplied in a single pass. Having requisitioned the machines, the scientists now also requisitioned a maintenance man—an IBM employee who had been drafted into the army. They were gaining adroitness at military procurement. The crates arrived two days before the repairman; in those two days Feynman and his colleagues managed to get the machines unpacked and assembled, after a fashion, with the help of nothing but a set of wiring blueprints. [6] 

Exxon donated its IBM CPC to UT in 1958. The CDC Cyber was essentially an updated CDC 6600. By the eighties, both UT and Exxon were in the ERA tradition with Cray hardware.

Is this Seymour Cray during installation of the CDC 6600 in 1966? Genuine question, as it does seem at least possible.

Official event for the CDC 6600 in 1966. David Young is at the very top center, somewhat less visible because of lack of contrast against the background. Curiously, Al Matsen seems not to be present in this photo. Both Young and Matsen are variously described as founders and heads of the UT Computation Center.

[1] SOR Successive Over-Relaxation and the work of David Young at TRW were prominently applied to fluid and heat flows for atmospheric reentry vehicles. At the same time, ADI Alternate Direction Implicit was being applied to fluid and heat flows for the oil industry in Houston. This was all in the mid and late fifties, and among the most important early applications of computers in modeling and simulation, alongside Dantzig's work on Simplex at RAND.

[2] For more about Exxon Houston, A Personal Retrospection of Reservoir Simulation, Donald Peaceman 

[3] The photos above related to the CDC 6600 are thanks to the Briscoe Center 

[4] Here's an excellent new post about the IBM CPC from Ken Shirriff.

[5] Notes about the post 6600 CDC hardware in Austin. UTCC upgraded to systems like the Cyber 170/750 and 175, which maintained compatibility with the older 6000 series code but introduced crucial magnetic-core and semiconductor memory enhancements to support operational runs on much larger scales. The Cyber 170/750 was a critical high-performance resource managed by the UT Computation Center. While the university's primary, general-access academic mainframes such as the original CDC 6600 and early Cyber systems were famously housed in an underground facility beneath the East Mall on the main campus, the Cyber 170/750 was deployed differently. The university utilized facilities at the Balcones Research Center, now known as the Pickle Research Campus, to host high-performance systems like the Cyber 170/750. At this off-campus site, the 170/750 was dedicated to handling specialized research and mathematically intensive data analysis tasks. Researchers also utilized a CDC Cyber 175 during the late seventies and eighties. At the time, the Cyber 175 was one of Control Data Corporation’s top-tier, high-speed scalar processors, and it was used intensively for complex finite element and alternating-direction method simulations to solve the types of convection-diffusion problems that frequently appear in reservoir engineering and geology.

[6] James Gleick, Genius: The Life and Science of Richard Feynman (New York: Pantheon Books, 1992)