The Sperry Rand Corporation
Amalgamating amalgams
This is the final entry in a four part series. You may be interested in reading about The ENIAC, which led to The Eckert-Mauchly Computer Corporation, which was then purchased by Remington-Rand.
Sperry Rand was formed through the merger of Sperry Corporation and Remington-Rand on the 30th of June in 1955. These two highly bureaucratic corporations were, themselves, built out of prior mergers, and it is therefore not at all surprising that Sperry Rand was among the largest of companies in the USA from the time of its formation. It had fifty four manufacturing centers spread across twenty one states, another in Puerto Rico, and another forty or so manufacturing facilities in another seventeen countries. The company manufactured a wide variety of goods including office equipment, agricultural equipment, aerospace equipment, naval equipment, and weapon systems.
At the time of the merger, it was quite well-known to the executives of both companies that they’d need to restructure and consolidate. This resulted in the new entity being comprised of just five primary divisions: Remington Rand, Sperry Gyroscope, Ford Instrument, Vickers, and New Holland Machine. This structure was later changed (around 1958) with the Remington Rand division being split into the Univac, Systems, Electric Shaver, Office Machines, and International. Later acquisitions would swell the size of all of these divisions.
The advancements made in computing during the second half of the 1950s are astounding. The first generation machines made by Eckert-Mauchly were built of vacuum tubes for logic and mercury delay lines for memory. The second generation made by the EMCC division of Remington-Rand used vacuum tubes for logic and magnetic drum for memory. The third generation made by Sperry Rand, started with the UNIVAC Solid State which used seven hundred transistors and three thousand magnetic amplifiers for logic and magnetic drum (with optional additional magnetic core) for memory. If we compare what this meant for the physical dimensions of these machines, the difference is easily visible. First generation machines could weigh up to thirty tons, occupy over a thousand square feet, and were only suitable for rather large institutions who could afford the machine and the building to house it. The UNIVAC Solid State machines weighed in under four tons and required just over five hundred square feet. Despite this size comparison, the UNIVAC Solid State machines had more memory, faster processing, used less power, and generated less heat. They were also radically more reliable. The UNIVAC Solid State was announced in December of 1958, and it was offered in four versions: SS I 80, SS I 90, SS II 80, SS II 90. In this listing, the SS denotes that the machine is solid state. The I vs II denotes the total memory capacity and peripheral set, and the 80 or 90 refers to the number of columns used by the cards. IBM’s cards were of eighty columns, and Remington Rand’s cards were of ninety columns. The I machines were the low-end of the line up with 5000 words of drum memory, and no tape drives. The II machines had an additional 1280 words of core memory and support for up to four magnetic tape drives. All of these machines included a card reader, card punch, line printer, and a ten-key for boot command entry and debugging. The usage of a ten-key is indicative of the type of programming typically done for UNIVAC SS - machine code. Of course, when speaking about any computer architecture, changes occur over time. An assembler was made, called X-6, which I am sure made everyone happier. As with other computers of this era, common routines were provided with the machine to speed software development. Another bit of software offered with the machine (though not immediately) was an IBM 650 emulator. The introductory price for a UNIVAC SS was $350,000, but it could be leased for $7000/month. Those prices in 2024 dollars would be around $3.8 million and $76,000 respectively.
An early win for the newly merged company was the AN/USQ-17 (also known as the UNIVAC M-460) designed by Seymour Cray and delivered to the Navy in March of 1958. It had a 30bit word and core memory of 32K words. This was the last machine Cray worked on before leaving Sperry Rand for Control Data. This system was redesigned to use silicon transistors allowing it to be more compact and fit into more secure housing. The redesigned version was then named AN/USQ-20 (alternately as CP-642, UNIVAC 1206, UNIVAC 1232, and G-40). A civilian version was made commercially available as the UNIVAC 490 (with updates designated 491 and 492). A multiprocessor version of that, the 494, followed later. This would be further updated in 1967 with ICs in a dual CPU version and reduced size. This update was designated AN/UYK-8. All of these machines were modular, real-time, and sold well.
Unfortunately, the mergers that created Sperry Rand made the UNIVAC SS slightly late to market. On the 5th of October in 1959, the IBM 1401 was released. The 1401 was, for the time, an insanely successful machine with over 12,000 installations. The 1401 was also cheaper than UNIVAC renting for $2500/month which would be around $27,000 in 2024 dollars. With the 1401 and all of the success of the IBM 650, IBM’s domination of the US mainframe industry was certain. IBM held roughly 71% of the market and Sperry Rand held just 10%.
It should not be thought that Sperry Rand was simply idle. Eckert and his team were hard at work on one of the world’s first supercomputers, the Livermore Advanced Research Computer, or UNIVAC LARC. This project was valuable enough and difficult enough that it made the company’s competitor to the 1401 regrettably late to market, but LARC is impressive when we consider the time of its design and manufacture. The design of the computer began in 1955 and was most certainly amended multiple times as more reliable transistors, cheaper core memory, and other serious hardware improvements were made.
This machine was big. At what is now Lawrence Livermore National Laboratory, a new building was constructed specifically to house this computer; that’s how big the thing was. The machine had four main cabinets with each being about twenty feet long, four feet wide, and seven feet tall. One cabinet was the I/O processor, another was the primary computer (think CPU), another two were memory. Each of the memory cabinets provided 16K of core memory. For storage, the machine had eight tape units, and for instructions it had a card punch and a card reader. LARC had teletype units, toggle switches, digital readouts, blinken lights, and buttons at the control console. LARC also had a sort of early RAM disk that became common on later UNIVACs. While tapes were used to load and store data, the system also had an array of twelve floating-head drums into which that tape data was loaded. These drums were four feet wide, three feet deep, and five feet tall. Each drum had a capacity of roughly 250,000 words where a word was twelve decimal digits. LARC weighed in at around 52 tons not including facility equipment, and the installed machine was smaller than what had been originally designed and offered. Initially, the machine was meant to offer two computers sharing a single I/O processor. It allowed for up to 24 drums of storage, and up to 40 tape drives. LARC was the fastest computer in the world for the years 1960 and 1961. It was then surpassed by the IBM 7030. LARC was also the first large scale use of printed circuit boards within a computer that I can find. However, in LARC, the PCBs were connected to one another by wires. So, as a result, the backplane had an actual wire for every single point-to-point connection within the computer, and these had to be the most direct and therefore shortest route possible as these thousands upon thousands of wires crossed one another’s path. The result was what appeared to be a mesh of wires that could reach a thickness of half a foot in places. Peering into this mesh with a fluoroscope (metal tube with a light and a magnifying glass at the end) for debugging or for the implementation of changes was risky to say nothing of actually performing a change as the density of wires would occasionally force a wire out of contact as the fluoroscope made its way through.
LARC was shipped out of Philadelphia in February of 1960 in five 18-wheelers. Three weeks later, it all arrived as did the forty person installation crew. The installation process took around two months. Another two months was spent getting things working after which nearly seven months of acceptance testing took place. The machine was then put into production use. A second LARC was delivered to the US Navy DTMB in May of 1960. LARC at Livermore was operational until December of 1968, while the Navy’s was operational until April of 1969.
In 1960, Sperry Rand purchased both the Adding Machine and the Cash Register divisions of Clary Corporation and folded their operations into the Office Machines division of Sparry Rand. In 1961, construction began on the Sperry-Rand building at 1290 Avenue of the Americas in Manhattan. The building was designed by Emery Roth & Son and was completed in 1963.
Following the delivery of LARC, the company’s Univac division needed to deliver its successor to the UNIVAC II, and it needed to do so quickly. UNIVAC III began in Philadelphia in 1958, but the work was transferred to the St. Paul location. Sperry Rand then made a bit of a partnership with Westinghouse. Sperry Rand made the computers, Westinghouse made the software and handled marketing and delivery. Westinghouse also made and supplied peripheral equipment and auxiliary control equipment.
UNIVAC III was loosely based upon many of the advancements made in LARC, but it was binary instead of decimal. It was fully transistorized, and it used core memory with each memory cabinet holding a maximum 16384 words where a word was 25 bits. The machine could be ordered with a half cabinet of RAM, a full single cabinet, one and half cabinets, or two full cabinets. Instructions were also 25 bits long. Like LARC, it offered drums for data storage populated from tapes. The system also had a card punch and a card reader for cards of either 80 columns or 90 columns, and it could punch 300 cards per minute or read 700 cards per minute. A high speed printer could output 700 lines per minute, and the computer could output punched paper tape as well. Two different operating systems were available for UNIVAC III, CHIEF or BOSS. UNIVAC III also had a standard assembler called SALT (Symbolic Assembly Language Translator), and COBOL was available for it. In a major departure from prior systems, UNIVAC III could load its OS, applications, and data from tape. In this case, the application would need to be at the start of the tape, and all data had to follow. The OS was aware of jobs, and job control data was usually stored separately. With tapes being the primary method of application and data loading, UNIVAC III could support up to 32 UNISERVO III tape drives, and an additional six UNISERVO II tape drives. UNIVAC III wasn’t instruction compatible with any previous UNIVAC systems, and this hurt it quite badly. The prices were slightly higher than those of UNIVAC SS, and with any buyer needing to rewrite his/her software anyway, why choose the UNIVAC III when one could purchase an IBM for less? UNIVAC III was released in June of 1962, and despite the UNIVAC III selling just 96 units, UNIVAC’s reputation was still fairly solid. The 1960 census was performed on UNIVAC, the Apollo missions maintained a 48kbps data connection via a UNIVAC on the ground, and governmental organizations were still using and installing UNIVAC.
The UNIVAC 1107 was introduced in October of 1962, and it was the first solid-state entry in the 1100 series. The 1107 could support up to 65536 words of 36bit core memory. This machine debuted (for civilian use-cases at least) a new type of memory developed by Sperry Rand. The CPU registers of the 1107 were made of thin-film memory. Thin-film is a form of magnetic storage, but it’s incredibly fast. Thin glass plates are covered with 4 micron dots (in this machine) of an iron-nickel alloy, referred to as permalloy, via vacuum evaporation and the use of a mask. With the surface completed, printed circuits are laid over the dots and glass to provide the sense lines. This technology provided access times of around 300 nanoseconds. Like other UNIVAC machines, the 1107 could make use of both UNISERVO IIA and UNISERVO III tape drives, and these could accept either metallic tape like UNIVAC I or mylar tape. This beastly machine could also make use of drum memory for storage with each drum unit capable of storing 300,000 36bit words. This machine was supplied with the EXEC I operating system, a Fortran IV compiler, and a macro assembler. This was still a big computer. Without any peripherals, UNIVAC 1107 weighed in at 2.4 tons and cost $888,750 which would be roughly $9.28 million in 2024 dollars. At this price, the machine would have been useless. The most basic build that would have been usable for real work would have cost the buyer $1,890,750 which would be close to $20 million in 2024 dollars. For this sum, the buyer would have received the CPU, a memory cabinet, card reader and controls, and a printer. This price would have included the necessary software, installation, and documentation, but nothing else. This kind of cost is precisely why most customers looking for computing power in this era rented rather than owned. While these machines were still more reliable than what had come before them, they required maintenance which meant skilled technicians. Most people would also have preferred tapes at this price point.
The UNIVAC 418 also known as the UNIVAC 1218 Military Computer saw its first deliveries in June of 1963. This was a fully transistorized machine using core memory. In dimensions, it was physically similar to the earlier AN/USQ-20 and had the same dual door front. Memory sizes ranged from 4K to 16K of 18bit words. This machine received an update in November 1964, and another in 1969. These updates modernized the machine’s memory and made the computers more reliable. The 418-III also increased the memory minimums and maximums to 32K and 131K respectively.
Announced in 1964 with the first deliveries having taken place in 1965, the UNIVAC 1108 was an awesome machine for its day. Comparing it to the 1107, CPU registers were now integrated circuits, and while the main memory was still magnetic core, it was far better made and faster. The computer supported multitasking and featured memory protection. This was followed a year later with the 1108A which could support three CPUs, four memory banks offering up to 262K of 38bit words, two programmable I/O controllers, and a math coprocessor called the UNIVAC Array Processor. About 296 UNIVAC 1108 series CPUs were sold. I am not certain how many total systems that represented. Most were shipped without the coprocessor. The 1108 was considered a success with that sales volume. These extremely large machines were never purchased in extremely large numbers due to their cost, size, power, and cooling requirements. UNIVAC had expected to sell just forty three 1108s. To provide a sense of who bought the big UNIVAC machines, customers for the 1108 included Lockheed, French National Railroad, Scottish National Engineering Lab, Boeing, Naval Ordinance Test Lab, NASA, University of Utah, Environmental Sciences Services Administration, Air France, Air Force Global Weather Central, US Census, Carnegie Mellon University, University of Maryland, and National Bureau of Standards.
A slimmed down version of the 1108, named the 1106, was released in December of 1969. This system was far cheaper and was effectively about half the speed of the 1108. While the 1106 sold better than the 1108, it wasn’t by a large margin.
In the mid 1960s, Sperry Rand began introducing terminals to replace the use of teletypes. This started with the Uniscope 100 in 1964. Of course, the UNISERVO units also improved. By 1970, the company was offering UNISERVO VIII-C tape drives holding 800 bits per inch.
Perhaps realizing that market was rapidly changing, Sperry Rand began producing IBM System/360 compatible machines. The UNIVAC 9200 and 9300 implemented the 16bit IBM 360 Model 20 instruction set, while the 9400 implemented the 32bit System/360 instruction set. The 9400, released in 1969, competed with the 360 Model 30. In 1971, the 9700 completed the line up with a machine that was twice as powerful as the System/360 Model 50 while being less expensive than the 370 Model 145. While running the System/360 instruction set, these machines still used UNIVAC peripherals like the UNISERVO tape drives, teletypes, terminals, and other equipment. The printer however was new for UNIVAC being of the IBM bar printer type. All of these machines shipped with FORTRAN, COBOL, and Algol. Thankfully, it would seem that the marketing department failures of EMCC and Remington Rand were not repeated. At least with the 9000 series, the Univac division of Sperry Rand emphasized time saving and labor saving as opposed to just rattling off how superior their machines were.
By 1969, Sperry Rand was feeling that the UNIVAC models 490 through 494 were burdens to be carried rather than assets. Specifically, the money being spent on software for the 494 was draining the company of resources that they felt could be better spent elsewhere. To this end, the company worked to develop a 494 emulator that would run on other UNIVAC models.
Starting in 1970, Sperry Rand secured a contract for computing platforms for the US Navy and Coast Guard. The first machine in that line was the AN/UYK-7. This was a 32bit system that could support both multiple CPUs and multiple I/O controllers sharing the same architecture in this regard as the UNIVAC 1108. This system was used for both ships and submarines. An airborne version of this was made, the UNIVAC 1832. In October 1984, this system was replaced by a modernized version using the same instruction set, the AN/UYK-43. Deliveries of this system continued until the year 2000. This updated version was quite reliable due to active redundancy, extensive self-tests, and on-line reconfigurability allowing it to run despite hardware failures.
Sperry Rand purchased RCA’s computer division in September of 1971. This RCA division had made the Spectra 70 series of computers that were (somewhat) compatible with and competed against the IBM System/360 line, and the RCA Series 2, 3, 6, and 7 which were (again somewhat) compatible with and competed against the System/370. These effectively replaced the 9000 series after 1972 and became the UNIVAC Series 70 machines. In October of 1973, the UNIVAC Series 90 was added to this line up to compete against the IBM System/370 Models 135 through 158. The 90/25 and 90/40 constituted the low end running OS/3, the 90/60 and 90/70 built out the middle running VS/9 which supported virtual memory. Another cheaper contender was added in June of 1974, the 90/30. The top end machine was the 90/80, but it wasn’t released until 1976. The Series 90 machines weren’t too successful as the IBM 4300 was a better buy in terms of price to performance, but the other machines sold well. With this acquisition and the customers whose maintenance and support contracts Sperry Rand consequently owned, the company enjoyed 40 consecutive quarters of profit.
Despite all of the most lucrative work being done in the lower-end of the market, the Univac division wasn’t going to forget their roots. The UNIVAC 1110 was introduced in 1972. This iteration of the 1100 series was fairly wild. All of the I/O was moved out of the CPU which was now termed the Command Arithmetic Unit. In the 1110, a single installation could now utilize six CAUs, four Input Output Access Units, up to 512K of plated-wire memory, and with more memory available via the older and slower core memory modules of the 1108 or 1106 systems (theoretical maximum of 16 million words). The majority of customers would never have used a maximum specification 1110. The majority were just two CAUs, one IOAU, and a modest mixing of memory units.
In 1975, Sperry Rand eliminated core memory from their high-end machines, and replaced it with semiconductor memory. The UNIVAC 1106 became the 1100/10. When you see these computers mentioned as something like 1100/12, that means the computer in question is a semiconductor memory 1106 with two CPUs. The 1108 became the 1100/20. The 1110 became the 1100/40. These machines were also unique in the history of UNIVAC for using a microcomputer during the boot process. The System Support Processor was a microcomputer responsible for loading microcode from 8-inch floppy disks into the UNIVAC’s CPUs and initiating the boot sequence. The final original 1100 series member was the 1100/80 introduced in 1979. A major feature of this computer was CPU and I/O processor cache (called a Storage Interface Unit, or SIU) that could be either 8K or 16K of 36bit words. The final 1100/80 systems could support four CAUs and four IOAUs. With this configuration, the SIU could bifurcate system resources provided into two dual CPU and dual I/O processor systems. Neat trick.
While the 1100/80 was being developed, the company was working on project Vanguard at their research center in Sudbury, Massachusetts. Unlike all previous UNIVAC systems, Vanguard would use the 4bit Motorola 10800 bit-slice processors along with custom ICs. This reduced costs so severely that the design team decided to duplicate the design and have the two processor implementations check each other, similarly to the BINAC and UNIVAC I. Despite the redundancy and error checking, the system still proved to be both cheaper and smaller than its predecessors. Vanguard became the 1100/60 upon its announcement to the world on the 5th of June in 1979, and the first delivery was made the following year. The 1100/60 replaced the 1100/10, 1100/20, and the 1100/40 in UNIVAC’s line up. This system was the first to ship with MAPPER (database and reporting software), and it included the 494 emulator mentioned earlier. The combination of MAPPER, the 494 emulator, reduced cost, reduced size, reduced power requirements, and increased performance brought the new system quite a bit of success. In its first year on the market, 528 processors were shipped. The 1100/70 released in 1981 was essentially the same system but with denser memory. Counting the sales of both the 1100/60 and the 1100/70, around 4000 systems were shipped. The UNIVAC 1100/80 was replaced by the 1100/90 in 1982 with the first units being delivered (as far I could find) in 1983. The 1100/90 was essentially an 1100/70 with liquid cooling and allowing four processor, four I/O units, and cache.
On the 16th of June in 1977, Sperry Rand purchased Varian Data Machines, previously Decision Control Inc. This was a maker of minicomputers based out of Newport Beach. By the time they became part of the Sperry Rand megacorp, they were selling the 620/L-100 and the V70 series with their primary competition having been the likes of the PDP-11 and DG Nova. A great testament to the quality of these minicomputers is a group of three Varian V72s that were still operational as recently as 2017 at the Bruce B nuclear plant in Ontario.
In 1978, Sperry Rand sold off several business units. At this point, the now decidedly computing focused company renamed itself to Sperry Corporation.
Sperry did okay in the early 1980s. Their mainframes were selling well, minicomputers were doing just well-enough to justify the continuance of the product line, but there was trouble. Some on the board realized that UNIVAC’s research was falling behind and their product lines were suffering for it. IBM was quickly taking more and more of the mainframe market, DEC was the king of the minicomputer market, and microcomputers were rapidly taking away market from both mainframes and minis. Sadly for Sperry, IBM had recently become king of the micros too! In the mid 1980s, this resulted in a plan to shift Sperry’s focus from hardware to software and services. What little hardware R&D in which the company was still engaged was split between AI focused LISP machines for which there was no immediate market, and mainframes that were a rapidly shrinking market. Of course, Sperry wasn’t the only mainframe company that was in trouble; GE and RCA had already left the market.
For one Werner Michael Blumenthal, allowing IBM to go unchallenged was unthinkable. Blumenthal had served as Secretary of the Treasury under the Carter administration, and he became vice chairman of Burroughs in 1980. In 1981, Blumenthal became chairman. In pursuit of his goal to challange Big Blue, he approached Sperry’s board in 1985 with a stock swap and was rejected with Sperry’s board regarding the offer as wholly inadequate. On the 13th of May of in 1986, he came back with another offer, also rejected, and Sperry countered with a higher dollar per share price. This was accepted and Burroughs acquired Sperry for $4.06 billion about half of which was cash and the other half a mixture of preferred stock and securities. The deal was closed on the 27th of May in 1986. Blumenthal spoke the next day saying:
We are trying to get over the fear factor - the fear that no one has the staying power against I.B.M. Once you get to a certain size - and no one has reached it yet - that fear tends to lessen.
Blumenthal became the CEO of the newly merged company, Unisys, and he vowed that he wouldn’t end of the mainframe computer lines of either company. This didn’t end up being entirely true, but the descendant of the UNIVAC 1100 series was the 2200 series, and then the ClearPath IX series, and then the ClearPath Forward Dorado series. The Exec 8 operating system that powered the last era of UNIVAC systems became OS 2200. The Burroughs systems are now ClearPath Forward Libra running MCP. Unisys began offering rather standard WinTel servers in the early 2000s, and in 2014, the company ended their proprietary mainframe architectures entirely. The mainframe line up became extremely high-end x86 systems with an introductory price of $325,000 and the highest end SKU commanding a price of $5.6 million. The reasoning for this shift is most easily seen in the company’s revenues. By 2014, nearly 86% of the company’s revenue came from services. Unisys, like the companies from which it was formed, continued to cater primarily to the enterprise computing and governmental sectors.
I have readers from many of the companies whose history I cover, and many of you were present for time periods I cover. A few of you are mentioned by name in my articles. All corrections to the record are welcome; feel free to leave a comment.
I have fond memories of using the Univac 1108 at the University of Maryland in the late 1960s. Perhaps you could add them to the list of buyers of the 1108? It ran Exec 8, which was very new and full of bugs. When I went to Bell Labs, they also had an 1108. Exec 8 in the early 1970s could also do time-sharing, but it was very inefficient. Something better came along: UNIX.