From the 1959 Eckert paper:
The primary function of the [I/O] Processor is the flexible, parallel, and coordinated control of all input-output equipment and transfers between this equipment and the memory system. The [high-speed arithmetic and logic] Computer is designed to perform rapid arithmetic computation with a minimum of interference.
The LARC Processor, however, may relieve the Computer of half of the work load it would otherwise have to bear. We thus have achieved as much as eight percent increase in speed for everyone percent increase in cost resulting from incorporating [a general-purpose instruction set] in the processor.
From the General Description UNIVAC-LARC System:
The code of the Computer does not include instructions for controlling input, output, and auxiliary storage operations directly. All such operations are handled by the Processor under executive command of the Computer. Consequently, the Computer is free to devote substantiall all of its time to performing arithmetic operations. Input, output and editing commands may be summarized by the Computer in well-defined pseudo orders having adjustable parameters and issued to the Processor singly or in groups via the core storage.
The Computer program is required only to place the summary orders in the storage, alert the Processor to their presence, and check for their completion after a period of time has passed during which the summary orders are executed independently by the Processor.
The Processor may also be programmed to relieve the Computer program of the task of generating and issuing summary orders and, in fact, relieve it of all concern with input, output and drum operations. Computer time can thus be devoted exclusively to arithmetic and related logical operations. In any case, the computer programmer will be concerned, not with the details of the input, output, and drum operations, but with logistical problems of anticipating storage requirements, allocating storage, and directing the flow of data to and from the storage.
From W. Lokke, "Early Computing and Its Impact on Lawrence Livermore National Laboratory," March 21, 2007, UCRL-TR-226840:
In September, 1955 the Laboratory signed a contract with Remington Rand for delivery in three years, 1958, of the $3 million LARC, the Livermore Advanced Research Computer.
Not fully transistorized, its tube driven core memory systems would provide 10,000 words a cabinet, with as many as ten cabinets being ddressable. The cost of memory, and its large size, kept the final version delivered to Livermore at three cabinets, or 30,000 words. A significant augmentation of memory was offered in the form of rotating drum memories, each drum holding 250,000 words. Livermore's twelve large drums were housed in a separate "clean" room needed to protect the delicate drums from contamination. The room's large picture windows showed off the red glow of the drum's spinning surfaces to theatrical effect. The machine's large vo lume required the construction of a 7,500 square foot building to house the LARC.
Manufacturing difficulties delayed delivery of the LARC for 27 months to June, 1960 - one month before the arrival of the first 7090. The LARC's performance, however, was as promised: it met or exceeded all of its specifications. Capable of 125,000 floating point multiplies per second, the LARC had become the world's most powerful computer. It was to hold the title for less than a year.
Despite its capability, the LARC was never to play its anticipated central role in weapons computing. Several factors combined to prevent this from occurring. As the machine's delivery delays increased, reliable code development planning for it became increasingly difficult. Fortran's success on the 704 and 709 in the late 1950s meant most code developers wanted Fortran on new machines. The impending delivery of the compatible 7090 would bring an effortless surge in computing capability to existing codes on the 709 and with two more 7090s in the pipeline, code developers were reluctant to return to the costly process of assembly language coding for an uncertain machine. Finally, internal difficulties meant negotiations with outside contractors for development of a Fortran for the LARC did not get underway until 1960.
See also Charles Cole, "The Remington Rand Univac LARC" for another description of the LARC at Livermore, including the difficulties of hardware upgrades involving rewiring the backplane.
[History page] [Mark's homepage] [CPSC homepage] [Clemson Univ. homepage]