# CRAY MUSEUM SELF-GUIDED TOUR

This museum was envisioned for a long time by Dean Roush and Les Davis. Dean collected most of the machines on display and several of the artifacts in the cabinet. He was a founder of Cray Research who provided mechanical engineering of computers for both CDC and Cray Research. Edna Bunn, one of the company's first employees, donated her husband Herm's collection, which added many more of the artifacts on display.

CRI plans to retain the first machine built of each major model. At this time, there are no plans to add any other machines developed by other manufacturers. Module types of other machines worked on by current and former employees are always welcome. We would especially appreciate any photos that show people at work or that depict significant historic events.

ERA/Univac items have been donated by Paramax and by several retired Unisys Corporation employees.

The Corporate Museum opened in July of 1990 with a reception for the Board of Directors. Early in 1995, the museum was moved from its original location in the Old Systems Building. The new building was sold in 1996, and the museum was shut down.

#### ( 1 ) Missing

Sorry, this section is missing from the adapted text.

#### ( 2 ) (mostly) Missing

...

Military computers that CRI founders worked on included Bomarc, Athena, Nike Zeus, Naval Tactical Data System (NTDS), and Tactical Army Computer System TACS). Univac and Unisys continue manufacturing the 1100 series of machines up to the present.

In 1954, transistors were coming into their own and Univac engineers were challenged to produce engineering models of machines to demonstrate the capabilities of machines using either transistors or magnetic amplifier technologies. As a result, Seymour Cray designed the Transtec 11 computer (see picture). This prototype did not result in a commercial machine, although its technology appeared in many subsequent military Univac machines.

In 1957, Bill Norris, Bill Kisch, Bill Keye, Seymour Cray, Frank Mullaney, Les Davis, Noel Stone, George Hanson, and Dolan Toth, among others, left Sperry Univac to form The Control Data Corporation (CDC) in Minneapolis, MN.

#### ( 3 ) ERA/UNIVAC Magnetic Drum

The first rotating mass storage to be designed was magnetic drums. In the early 1950s, ERA/Univac designed several models. An example of one of the largest models is on display. Machines such as the ERA ATLAS (1101) and UNIVAC 1103 used drums for direct access memory.

ERA/Univac worked closely with 3M to develop magnetic coatings that were later used on disk storage units. ERA/Univac and IBM jointly developed floating-head technology that was used on later models of drums prior to its use on disks. Drums continued to be used into the mid 1960s but were eventually completely replaced by disk units. Control Data developed an operating system called Drum Scope that was used on its CDC 3600/3800 computer systems.

#### ( 4 ) CDC 1604 Computer

The first machine produced by Control Data Corporation (CDC) in Minneapolis, MN was the CDC 1604. This is one of the first solid state computers and was the most powerful computer in its day. Applications included real- time data processing, weapon-system control, solution of large-scale scientific problems, engineering and commercial applications, and industrial processes control. The 1604 on display came from the University of Illinois Coordinated Science Lab. It was one of the first computers used for teaching computer science courses. This is also one of the first machines to feature compilers and "standard software." Note how the chassis are arranged in "pages."

The 1604 had a clock speed of 5 microseconds Peripheral devices included paper and magnetic tape, teletype machines, punched cards, and line printers.

Notice that there are switches on the console. Since software was very primitive and usually loaded from loaded from paper tape, a small degree of program control was available through sense switches. For many years, ANSI standards for Fortran required compilers to include features to accommodate these physical switches.

The 160A, announced in 1962, even though it sold for about $500,000 fully equipped, might be considered one of the first "personal computers." It uses the same basic technology as the 1604 but was designed for dedicated applications such as production control. Examples of early applications are control of Linotypes and lathes. The 160A on display was first owned by the U.S. Air Force and was later acquired as a behavioral science tool by Dr. Karl Smith at the University of Wisconsin Behavioral Cybernetics laboratory in Madison, Wisconsin. Dr. Smith noted that he used it for the first real-time analog-to-digital-to-analog system to study motor sensory feed back control. He claimed that his experiments to invert, reverse, and delay the retinal feedback of eye movements discredited and negated a century of superficial research. He considered the CDC 160A to be "the most outstanding scientific device ever built." The teletype shown alongside the 160A served as a console. Seymour Cray noted in a 1975 address that he designed the 160A in I week. Programmers Bill Glass and Garner McCrossen developed an early Fortran compiler for the 160A. #### ( 6 ) Seymour R. Cray Jr's Patents Many of the patents that Seymour Cray, Jr was awarded during his years with Control Data Corporation and Cray Research are shown on the wall behind the CDC 160A. Usually, patent rights belong to the company for which an engineer works and it was the tradition to give each engineer who earns a patent a framed copy with a token one dollar bill. The desk in front of the display is one that Mr. Cray used while at Cray Research, Inc. #### ( 7 ) CDC 6000 Series Computers In 1962, Seymour Cray moved back to the Chippewa Falls area and opened a CDC Laboratory in Hallie where he led the designed of the CDC 6600 computer. This machine, which was announced in 1964, sold for around$7 million and was the first computer designed in the Chippewa Falls area. Serial No. 1 was delivered to Lawrence Radiation Laboratory in 1964. The 6600 is believed to have been the first computer to be designated as a "supercomputer." It was used for many scientific applications, such as weather forecasting, climate analysis, nuclear analysis, and automobile, aeronautical, and missile design. The 6600 featured parallel functional units and used 10 peripheral processors (PPUs) for distributed processing. It sported the fastest clock speed for its day (100 nanoseconds). The 6600 was the first commercial computer to use a CRT console; CRTs and radar screens had been used on earlier machines. CDC checkout engineers created computer games such as Baseball, Lunar Lander, and Space Wars, which became incentives for getting the machines operational. These are thought to be the first computer games that used monitors.

Several features of the 6600 were revolutionary. First, it employed independently operating functional units that enabled up to 10 machine instructions to be processed simultaneously; an instruction stack further isolated the CPU from slow memory accesses. Second, in addition to a central processing unit, it featured 10 peripheral processors and 12 data channels that allowed the CPU to perform arithmetic operations asynchronously from input/output operations. Third, an innovative Exchange Jump instruction enabled the peripheral processors to interrupt the central processor and exchange its currently operating program with another program from central memory. A 16-word control package that was loaded as part of the exchange ump operation provided information necessary for executing the exchanged program. This allowed a number of programs to share the central processor, executing in turn other programs waiting for completion of slow data transfers to or from peripheral devices. Fourth, the CDC 6600 was one of the first computers to use R-22 (Freon) refrigerant cooling.

The 6500 on display was acquired from Purdue University when it was removed from service in June of 1989. A 6500 consists of two 6400's. Purdue originally ordered a single 6400, which is a cost-reduced 6600 and lacks the functional units; by delivery time, it had been upgraded to two processors, which accounts for the 6400 on one of the bays. Purdue's 6500 is thought to be the last operational 6000 Series machine; one of its last applications was to generate the payroll for the university.

In the software area, while it was not unusual for many customers to develop their own operating systems, most 6000 Series customers used the standard CDC offerings: the SCOPE Operating System or the KRONOS Operating System. These were later renamed NOS/BE and NOS, respectively. A predecessor of KRONOS was MACE, which was never an official CDC product. Personnel in Chippewa Falls developed an early operating system called the Chippewa Operating System which was used as a basis for MACE. Garner McCrossen developed early Fortrun cormpilers for the 6000 Series computers. These compilers were innovative in that they took advantage of the unique architecture of the machine, allowing CDC to win many government contracts with its superior benchmarking results. Fortran has remained the primary scientific language for engineers and scientists.

#### ( 8 ) Input/output Devices

A supercomputer cannot operate efficiently unless it is supported by high-performance peripherals that can supply data and accept data at speeds that keep pace with the arithmetic processes. Early devices included punched paper and mylar tape, magnetic tape, punched cards, magnetic drums, and disk units.

Control Data very early in its history decided to develop its own peripheral devices and became a highly respected manufacturer of paper and magnetic tape units, card punches and readers, optical character readers, disk units, and printers. Their peripheral division is today owned by SeaGate.

Paper Tape

Both the CDC 1604 computer and CDC 160A computer were equipped with paper tape readers and punches. Paper tape was mostly used for loading start-up (bootstrap) programs on these computers.

The Flexowriter, shown here, was a device used during the 1950s and 1960s to prepare punched tape programs and data for entry into the computer and for printing punched tape output generated by the computer. The Flexowriter was "off-line," which means that the device was not connected directly to the computer because it was too slow and could not keep pace with with the arithmetic processes. Also, printers available at the time produced very crude output, while the Flexowriter, being a modified typewriter, produced letter quality output.

The Model 35 Teletype shown with the CDC 160A was a punched tape device commonly used online and served as an operator workstation by providing both keyboard input and typed output.

Punched Cards

Businesses and financial institutions used punched cards long before the advent of computers. While card stock was standardized, two competitive punch systems were in use. IBM favored the 80-column Hollerith card while Remington Rand tabulator equipment in use during the 1940s and 1950s used a 90-column punch with two 45 rows of 6-bit characters. The punches were round rather than rectangular. Unlike the IBM card punch which punched each character as it was typed, the Remington Rand card punch featured a punch set-up so that the operator could preset the entire card and then punch it. This was thought to be both faster and more accurate than the IBM's conventional 80-column punch. Remington Rand Univac took a stand to support the 90-column card and not support the 80-column card. However, the 80- column format dominated the industry, and the 90-column card became extinct. Seymour Cray, when asked why Univac Corp. fell behind IBM and CDC in later years, replied "round holes."

The Cray Research CRAY-1 computer system Maintenance Control Unit (MCU) was equipped with an 80- column card reader and 80-column card punch, which were the primary input/out tools for programmers. These are displayed with the CRAY-1 Serial No. 1 Computer System.

Printers

The first printers used on-line (that is, directly connected to computers) were large slow devices. There were several types of mechanisms used for these "Impact" printers. Typically, a typecast character or line of characters would be mechanically positioned to be struck by a hammer mechanism. Ink on a carbon ribbon would be transferred to continuous form computer paper that was pulled or driven by a "tractor" mechanism.. The CDC Model 501 line printer, shown here, printed a line at a rate of about 100 lines per minute(?); there was a character wheel for each character position; lines of print often came out wavy because the wheels were difficult to keep in alignment. The character set was limited to capital letters, numbers, and a few business symbols.

The Model 501 printer was replaced in the late 1960s by a Model 512 printer. A "train" or "chain" of characters was positioned so that characters were printed one at a time. The printer featured a larger character set that included both upper and lower case letters and a much faster print rate so that copy seemed to spew forth from the printers at an impressive rate.

The first printer used with the Cray Research CRAY-1 computer system was a Gould printer that was connected to the Maintenance Control Unit (MCU). This printer used a toner and photosensitive paper such as used in copiers of the day.

Printers advanced through many technological phases, such as ink jet, xerography, and laser technologies beyond the scope of this museum.

Magnetic Tapes

Magnetic tapes were the most common method for storing large amounts of data on-line on early computers. They provided a reliable, fast method of both storage and retrieval. Data on reels could be stored off-line in large tape vaults. Because Mylar tapes were very weak, Univac Corp. used a steel tape for many of its military computers (see example on display in small cabinet). Magnetic tapes and the units that were used for recording and reading them dramatically improved between the 1950 and 1970s. Stronger Mylar, better magnetic coatings that allowed denser data recordings, and faster, more accurate tape drives all contributed to tape technology becoming the most popular storage technology for about 25 years.

The tape units displayed with the CDC 1604 were typical of those in use in the 1950s.

Tape units, while faster than punched cards, were still unable to match the speeds of the central processing units. One primary disadvantage is that they are serial devices, that is, information on tapes is in one long stream. Positioning a tape in order to read/or write specific data often took longer than reading/or writing the data.

Disk Storage

Disks, as a recording medium for computers, were under development for many years. The primary advantage with disks over tape is that they provide for randomly accessing data. The mid 1960s saw the development of extremely large disk storage units using platters up to nearly 6 ft. in diameter. The largest disk on display is approximately 3 feet in diameter. Up to 19 of these disks rotated vertically in a cabinet called the Bryant Disk File. These units, none of which is believed to be in existance had such high angular momentum that they bolted to the floor so that they wouldn't "walk away." Control Data Corporation produced the Model 6638 Disk Unit for use with the CDC 6600 computer and the Model 7638 Disk Unit for use with the CDC 7600 computer that featured two spindles of these horizontally mounted disks. The spindles rotated the disks in opposite directions to cancel the angular velocities. Smaller versions of disk units became possible when better magnetic coatings and read/write mechanisms allowed higher recording densities. These units were soon replaced by units that featured removable disk packs that allowed offline storage of data, similar to that provided for tape.

Cray Research, Inc. does not produce its own disks but purchases them from Other Equipment Manufacturers (OEM). The disk units are then marketed with a Cray Research name and model number. The following disks have been offered:

Model Machines supported OEM supplier
DD-19CRAY-1 Control Data Corporation
DD-29CRAY-1 Control Data Corporation
DD-49CRAY X-MP; CRAY-2
DD-39CRAY X-MP Fujitsu of America
DD-40
Ibis
DD-50
Sea Gate

#### ( 10 ) CDC 7600 Computer

Upon completion of the design for the 6600 and hand assembly of the first five machines, the team in Chippewa Falls went on to design the CDC 7600. An interim machine, the CDC 6800 (see artist's rendering), was designed but was never completed, although its cabinetry and other features were adopted for the 7600. The CDC 7600 was announced in 1969. It was sold to many of the same customers as the 6000 Series machines. Serial No. 3, on display, was assembled in Chippewa Falls and shipped to the Los Alamos Scientific Laboratory in 1970 for use in nuclear research.

The 7600 featured a two-level memory (Large Core Memory or LCM and Small Core Memory or SCM). Examples of these memory matrices are in the display cabinet. The clock speed for the 7600 was 27.5 nanoseconds. About forty 7600 computers were eventually sold before CDC replaced the 7600 with the CYBER series.

The first operating system for the 7600 was SCOPE 1. This was later replaced by SCOPE 2, which was subsequently renamed NOS/BE. It ran a variety of Fortran compilers and the CDC COMPASS assembler. Customers included the U.S. government, univesities, scientific laboratories, and private industry such as automobile and aeronautical manufacturers.

Sample of modules, memory planes, and cold bars are on display in the large artifacts cabinets.

#### ( 11 ) Control Data 7611 I/O Station

Supporting the CDC 7600 is a 7611 I/O Station. It consists of six Peripheral Processors (PPUs) and served the same function as served by the Cray Research I/O subsystem, namely gathering data for input to the central processor and distributing output. The 7611 on display came from the U.K. Additionally, two 7612 systems were built and were used within Cray Research by design engineers and technical publications but were never offered commercially. The 7612 was called the "phone booth" because the unit consisted of two 761 1 chassis joined by a memory chassis in a matching cabinet to form a U-shaped cabinet. Seymour Cray used a 7612 to design the CDC 8600 computer. Neither 7612 is believed to be in existence, today.

#### ( 12 ) CDC 8600 Computer

Having completed the 7600 and turned it over to the Control Data Arden Hills Operations for mass production, the Chippewa Falls team headed by Seymour Cray went on to design the CDC 8600 computer. This scalar machine (see picture) was never completed, although many feel its feasibility was proven before the project was terminated when CDC closed its laboratory in Chippewa Falls in 1973. The 8600 featured multilayer boards similar to those later used in the CRAY-2 system, The 8600 had pins in the Z-plane, that is, vertically from board to board; it was the first CDC machine to use an 8-bit ASCII code and a 64-bit word. Its clock speed was to have been 8 ns. Although some of the modules have survived (see display cabinet), the chassis is reported to have been scrapped in 1982.

#### ( 13 ) CRAY-1 A/B Computer Systems

In 1972, former CDC founders, Seymour Cray, Frank Mullaney, George Hanson, and Noel Stone, formed Cray Research. The new company, which established its headquarters in Chippewa Falls, began its operations with engineering teams headed by Les Davis, Dean Roush, and Harry Runke.

Seymour Cray, as computer architect, provided the technical vision of a CRAY-1 computer that was twice as fast as the CDC 7600 (12.5 ns for the CRAY-1 system vs 25 ns for the 7600 system) and demonstrated balanced scalar and vector performance. The computer was also innovative in its use of reciprocal approximation for division. Harry Runkel was the chief logic designer; the machine was built primarily from nine off-the-shelf LSI circuits,

which considerably shortened the development cycle. Seymour Cray, Harry Runkel, and Les Davis wrote about 1,300 pages of Boolean algebra to describe the approximately 100 different modules in the system. However, one of the biggest challenges was cooling the system; here, Dean Roush provided the mechanical engineering know- how to develop the cabinetry and cooling technology, for which he holds the patent.

Other technical staff who helped design and develop the CRAY-1 system included Gerry Brost, Roger Brown, Larry Gullicksen, George Leedom, Jim Mandelert, Rick Pribnow, Al Schiffleger, Dave Schulist, Ken Seufzer, and Jack Williams.

Serial No. 1 of the CRAY-1 computer system, the company's first product, was powered on in May of 1975 and officially introduced in 1976. Serial No. 1 was shipped to The Los Alamos National Laboratory in 1976 for a 6-month trial period. At about the same time, the National Center for Atmospheric Research (NCAR) persuaded Cray Research to add error correction to the CRAY-1 system and to begin the development of standard software. NCAR was Cray Research's first official customer. The company became profitable upon the acceptance of SN 3 by NCAR in December of 1977 (see photo). Because of the addition of error correction, SN 1 of the CRAY-1 system that is on display is 4 inches shorter and has 8 fewer modules per chassis than all other CRAY-1 computers and all CRAY X-MP computers. SN 1 was retired from service in the U.K. in May of 1989. See framed history.

CRAY-1 computers outperformed all other machines in their time and even today are recognized as having set the standard for the supercomputing industry. Applications of the CRAY-1 computer system include climate analysis, weather forecasting, military uses, seismology and reservoir modeling for the petroleum industry, automotive engineering, aeronautical engineering, space engineering, computerized graphics for the film industry, and so on.

Features of the CRAY-1 system include a fast clock (12.5 ns), 64 vector registers, and 1 million 64-bit words of high speed memory. Its revolutionary throughput figures -- 80 million floating-point operations per second (Mflops) -- were attributed to the balanced vector and scalar architectures and to high-performance software that was tailored to efficiently use the machine's architecture.

The 5-ton mainframe occupies only about 30 square feet. This extremely dense packaging challenged the engineers responsible for removing heat generated by the modules. Thus, one of the most significant engineering accomplishments was the cooling system. Liquid coolant (Freon) is pumped through pipes in aluminum bars in the chassis. The copper plate sandwiched between the two 6 x 8 inch circuit boards in each module transfers heat from the module to the aluminum bars. The power supplies and coolant distribution system were craftily placed in seats at the base of the chassis. These seats were useful for checkout engineers who would sit on them when attaching oscilloscopes to test points on the modules. Examples of the modules and cold bars are in the display case. The CRAY-1 system contains about 200,000 integrated circuits, and its wire mat contains approximately 67 miles of wire. The first CRAY-1 computers each took a year to assemble and check out because of the large amount of hand wiring and assembly required. A continuous stream of engineering improvements and the customization of the machines to satisfy unique customer requirements meant that no two CRAY-1 computers were identical.

The model designations A, B, and C, refer to the sizes of memory: 1 million words, 500 thousand words, and 250 thousand words, respectively. Although l7 CRAY-1A/B computers were built, no CRAY-1C computers were ever built. None of the original CRAY-1 systems are in service today; many of them have been retired to museums, including SN 14 at the Smithsonian Institute in Washington, D.C.

The requirement for standard software was a huge challenge. The company was faced with the need to deliver software with virtually no time, tools, or staff with which to develop it. One option under consideration was to contract out the effort. However, the programmers already hired, including Margaret Loftus who had been hired in May of 1976, prevailed on management and with the support of George Hanson, began developing COS Version 1. Staff was expanded and software analysts began the effort, building on a nucleus of code for EXEC and STP developed by Bob Allen in Chippewa Falls. In 1978, the first standard software package consisting of the Cray Operating System (COS), the first automatically vectorizing Fortran compiler (CFT), and the Cray Assembler Language (CAL) was officially introduced. Dick Nelson, who is the company's first programmer analyst, is responsible for the design and development of CFT, the world's first automatically vectorizing Fortran compiler.

In 1979, the CRAY-1S (see photo) was announced. It featured many different models, increased memory sizes (up to 4 million words), additional functional units, and an I/O subsystem (optional). Over the next three years, thirty-nine CRAY-1S computers were built and sold. These machines were some of the first to be used by commercial engineers for fossil fuel research, reservoir modeling, and automotive and aerospace applications.

In September of 1982, the CRAY-1M system, the last of the CRAY-1 series of computer systems was announced. The first CRAY-1M system (see picture) was saved and is on display in the CRI Customer Services Building in Chippewa Falls. It is a single-processor unit that used the same memory technology as the CRAY X-MP system. It is significant because the cost of producing a CRAY-1 system was dramatically reduced with this model, and it was sometimes referred to as the "two-for-one sale." However, only nine CRAY-1M systems were built before they were replaced in the product line by single processor CRAY X-MP systems, which featured the same high- performance CPU as the rest of the CRAY X-MP series. The CRAY-1M system in Saudi Arabia was the first Cray computer to operate for a year without a hardware failure. Serial number 8 of the CRAY-1M system is on display at the Computer Museum in Boston, MA.

A uniprocessor.
Vector processor, uses pipelining and chaining to gain speed.
12.5-nsec clock. Fast scalar.
Uses only four chip types with 2 gates per chip.
64-bit word size up to 4 Mwords of storage.

The CRAY 1-S has bipolar (in units of 4K RAM), and the newer (1982) CRAY 1-M has MOS memory (in units of 16K RAM).

Logic chips - ECL with a gate delay of .7 nsec.
Main memory banked up to 16 ways. The bank busy time is 50 nsec (70 nsec on the 1-M) and the memory access time (latency) is 12 clocks (150 nsec).
No virtual memory
Register-to-register machine
8 registers of length 64 (64-bit) words each
No half precision.
Double precision (128 bits) is through software and is extremely slow (factors of about fifty times single precision (64 bits) are common).

There is only one pipe from memory-to-vector registers, resulting in a major bottleneck with loads and stores to memory from registers. Loads can be chained with arithmetic operations; stores cannot.

Low vector start-up times and fast scalar performance make this a very general-purpose machine. Max. performance 160 Mflops; 64-bit arithmetic; max. attainable sustained performance 150 Mflops. There are codes for matrix multiplication and the solution of equations which get close to this. Maximum scalar rate is 80 mips. It is easy to attain over 100 Mflops for certain problems, even using Fortran.

#### ( 14 ) CRAY-1 Maintenance Control Unit (MCU)

(being written by Rick Pribnow)

[Come on, Rick, what's taking so long? :-)]

#### ( 15 ) CRAY I/O Subsystem (IOS)

A CRAY I/O subsystem (IOS) is on display with the CRAY X-MP computer system. This device accumulates input for the central processor and distributes output, which frees the CPU from time-consuming peripheral activities and provides connectivity with a variety of other vendor's equipment, such as CDC, IBM, VAX, Sun, etc.

The capability to move information to and from the mainframe efficiently is imperative for a high-performance computer. It requires not only exceptionally high channel speeds but reliable and high-capacity secondary storage and peripheral devices. Cray Research has always paid close attention to these needs and has led the industry in high-speed channel development.

The IOS consists of up to four independent peripheral processing units (IOPs). Originally called A processors, these units were once intended to be used in clusters for array processing in an early architectural scheme planned for the CRAY-2 system. The cost of producing them was higher than anticipated, which made the overall cost of an array processor system infeasible, so the idea was abandoned.

Although the IOS was first offered on larger models of CRAY-1 S systems, it wasn't until the CRAY X-MP system was introduced that an IOS became an integral part of a CRI supercomputer system. On early CRAY-Y- MP computing systems, the same model IOS that was used on the CRAY X-MP system was attached to the CRAY Y-MP CPU cabinet. This made one arm of the distinctive Y shape; the other arm was an SSD solid-state storage device. On later CRAY Y-MP computers, one or more IOS sytems were built into the mainframe cabinet.

#### ( 17 ) CRAY X-MP Computer Systems

In 1982, Cray Research announced the CRAY X-MP, its first multiprocessor computer. With their 9.5 ns clocks, the two CPUs for the CRAY X-MP system were significantly faster than those for the CRAY- 1 system. The IOS, having proven its worth, became a standard feature. SN 101, the first CRAY X-MP system built, is on display. The first 4-processor version (see photo), fondly known as Abner, was originally blue with velvet cushions. One of the cushions is on display. These blue skins and cushions were replaced by red vinyl skins and cushions when the system was sent to the Cray Research Computer Center in Mendota Heights, MN.

Replaced in the CRI product line 1988 by the CRAY Y-MP series of computer systems, the CRAY X-MP series of computer systems was one of Cray Research's most popular products to date, with 189 produced.

Operating systems used on CRAY X-MP systems include COS; CTSS (the Cray Time-Sharing System developed by customers at the Lawrence Livermore Laboratory and the Los Alamos National Laboratory), and UNICOS based on UNIX System V technology.

This multiprocessor pipelined vector machine has the same architecture as the CRAY-1. The major difference is that there are three paths from memory to the vector registers, and the clock cycle time is 8.5 nsec on all machines shipped after August 1986 (machines built before August have a cycle time of 9.5 nsec.)

The current machines come with 1, 2, or 4 processors. Gather/scatter hardware is available on the 2- or 4-processor version of the machine. The gather/scatter can be chained to a load/store operation. Users can control all processors through calls in Fortran. The processors share memory.

Main memory - ECL 4K RAMs with 25-nsec access time. (Interleaving to 64 banks is possible.)

Memory up to 16 M (64-bit) words X-MP-2 MOS. (Bank busy time is 68 nsec and memory access time is 17 clocks.)

X-MP-4 ECL. (Bank busy time is 34 nsec and memory access time is 14 clocks.)

ECL logic with .35-.5 nsec gate delay and 16 gates/chip.

High-speed connection at 1024 Mbytes/sec per channel (max. 2) to a CRAY SSD. The SSD comes in various sizes up to 512 Mwords of secondary MOS memory. Data transfer to high speed (1200 Mbyte) DD-49 disk takes 10 Mbytes/sec. Recent peripheral enhancements as reported under the CRAY-2.

Peak of 235 Mflops per processor.

There are many possible front ends including IBM, CDC, VAX, and Apollo.

Delivery: Announced in August 1982, first system delivered in June 1983.

#### ( 18 ) Cray Research SSD Solid-state Storage Device

Shown with the first CRAY X-MP system is the first SSD solid-state storage sevice, which was announced in 1982. This unit was designed for rapid secondary storage to provide nearly immediate reading and writing of large data files. The channel used with it to communicate with the CPU offered data transfer rates of up to 1250 million data bytes per second, far exceeding any other data transfer devices in its time. SSDs were offered in sizes of 64, 128, or 256 million bytes of storage. The hole in the cabinet of the CRAY X-MP was to attach a very high speed (VHISP) data channel to an SSD. The link was referred to as the "skyway." SSD SN 1 did not accommodate this link. Note how the cabinetry of CRAY X-MP systems can be differentiated from CRAY-1 systems and early SSD systems.

Through use of software, the SSD is logically accessed as if it were a disk unit. For this reason, while it was under development, the SSD was referred to as the solid-state disk unit.

Today, the SSD is an integral part of CRI computer systems and is included in the mainframe cabinet.

#### ( 19 ) CRAY-2 Computer System

The first CRAY-2 system built, SN 2001, is known as "Bubbles." It was returned from the Lawrence Livermore Laboratories Magnetic Fusion Energy laboratory in 1993, where it was shipped in 1985.

The CRAY-2 system was the first computer to have modules immersed directly in a liquid fluorocarbon coolant, which maintains the computer at a constant temperature of 70 degrees Fahrenheit. The coolant, called Huoninert liquid was sometimes used as a human blood substitute during surgery. In CRAY-2 systems, the coolant flowed directly over the modules and was pumped to a heat exchanger. The reservoir was used for cycling the coolant and removing the coolant for maintenance. Later models had a scaled reservoir. A bottle of water and a bottle of fluorcarbon are included with the display so that visitors can note that while they appear the same, fluorocarbon liquid is much heavier than water.

In addition to its innovative cooling technology, some of the significant features of the CRAY-2 system were its extremely high speed clock (4.1 ns), 4 central processors, and its memory size -- up to 512 million words or 4 billion bytes of immediately accessible main memory.

Twenty nine 4-processor CRAY-2 systems were built before the line was discontinued. In addition, two one- processor prototype systems were built and one 8-processor system was built.

The CRAY-2 system is the first Cray Research computer and first large-scale mainframe computer to feature an operating system (UNICOS) based on the UNIX System V technology. Portability was one of the foremost reasons for choosing a system based on UNIX because of the different architectures used for the CRAY X-MP and CRAY-2 systems. Also, hardware performance was not to be compromised by software. The equity that Cray Research and customers had in software needed to be protected without restricting the creativity of design engineers. Up until this time, customers had always assumed that they or the vendor would redo an operating system whenever a new supercomputer such as the CRAY-2 system was introduced. Not only was this practice increasingly expensive as the size and complexity of operating systems grew, but also the development time required delayed the efficient use of new hardware by 3 to 5 years.

This is a 4-processor (quadrant) vector machine with pipelining and overlapping but no chaining. There are more segments in the pipes than in the other CRAYs. Multitasking primitives have same syntax as the X-MP.

The system has a 4.1-nsec clock cycle time.

Overheads for vector operations are large:

22 cycles for vector multiply
63 cycles for vector store

Memory is 256 Mwords of 256 K DRAM in 128 banks. The bank busy time is 57 clocks, and the scalar memory access time is 59 clocks. Local memory is 16 Kwords, 4 clocks from local memory to vector registers. Vector references from local memory must be with unit stride. There are 8 vector registers each with 64 elements.

Peak performance is 488 Mflops per processor. A matrix multiply code has run at 1.7Gflops on 4 processors.

Software includes: UNIX-based OS (called UNICOS); C compiler; CFT2 (Fortran compiler); CFT77.

Cray has an ongoing commitment to high-speed peripherals and fast network links. HSX is a 100 Mbytes/sec link for connecting CRAYs together. CRAYs can be linked to Ultra Corporations 1.6Gbit Ultra bus in addition to standard connections with Ethernet (TCP/IP), and VME buses. The DD-40 disks each hold 5 Gbytes and have a transfer rate of 10 Mbytes/sec.

Delivered: NMFECC, NASA Ames, University of Minnesota, Harwell Laboratory, Stuttgart, and Ecole Polytechnique (Paris).

#### ( 20 ) CRAY Y-MP Computer Systems

Models include: Y-MP 2E, 4E, 8E, 8I

This is a multiprocessor pipelined vector machine. It has a similar architecture to the CRAY X-MP. A major difference is the availability of 32- as well as 24-bit addressing. The cycle time is 6 nsec, and it is an 8 processor machine. As in the X-MP there are three paths from memory to the vector registers.

There are only three module types, for the CPU (8 modules), the memory (32 modules, with 1 Mword/module), and the clock (1 module), making 41 modules in all compared with 144 in the X-MP. Each module is on an 11" by 21.2" board. 2.5 $\mu$ ECL in 2500 gate arrays with gate delay of 350 picoseconds. There are 312 arrays per processor on four PCBs with a power dissipation of 9 Watts per array.

Architecture up to a maximum of 2GB central shared-memory (in largest processor configuration only).

The processors share a common memory of 32 Mword in bipolar SRAM with a 15 nsec access time and a bank busy time of 102 nsec. Memory is interleaved in 256 banks. Total bandwidth is 340 Gbytes/sec (32 words/CP per processor). The Y-MP comes with a 128 Mword SSD as a standard feature. 2 IOSs can be fitted, each with a 4 Mword buffer memory.

Peak performance of 4 Gflops. Overall performance of about 30 times a CRAY-1. Each processor should outperform a single X-MP processor by a factor of 1.4 in vector mode (1.2 in scalar).

Data Transfer up to a maximum of 10.4GB/s IO bandwidth (in largest processor configuration only).

Performance over 2Gflops sustained performance claimed for full 8-processor Y-MP/8, with 2.13Gflops achieved on SLALOM benchmark! Single CPU Y-MP/8 SLALOM performance measured as 0.28Gflops.

Programming Environment Extensive parallel libraries and kernels, with compilers featuring autodetection and replacement of BLAS1/2/3, LAPACK and FFTLIB routines. CDBX debugger, along with tools for performance monitoring, analysis, enhancement and visualisation, provide a comprehensive programmer support environment. Network interfacing provided via TCP/IP, NFS, HiPPI, FDDI, etc. Graphics support for AVS, Explorer, X11 and Cray Visualisation Toolkit, etc. Autotasking Expert System proprietary automatic parallel multitasking tool.

COS and UNICOS are both supported. In addition to CFT77 a new Fortran compiler will be available. Performance tools include dynamic and static analysis, tuning, and debugging aids. Can run in X or Y mode, and software can also run (in X mode) on the X-MP.

Software is fully binary-compatible throughout Y-MP range (including C90 and EL systems).

Languages new compiling system'' releases of sophisticated vectoristing and parallelising compilers for Cray Fortran-77, C and Ada (using Cray extensions and directives) together with existing Pascal and Common Lisp products. Supports IEEE floating-point and Fortran-90 array syntax, along with extensions for variable-length arrays and restricted pointers. Manual multitasking through (local) microtasking and (global) macrotasking directives distributed across up to 16 CPUs. Scalar processing supported via inlining, overlapping instructions and operation grouping; vector processing supported via pipelined overlap, vector chaining and multiple pipelines.

There are many possible front ends including IBM, CDC, VAX, and Apollo. There are four VHISP channels each rated at 1250 Mbytes/sec, eight 100 Mbytes/sec HISP channels, and eight LOSP 6 Mbytes/sec channels. A full range of disks, tapes, terminals, workstations, and networks (including TCP/IP) are supported.

Cooled using inert fluorocarbon but not with liquid immersion technology (as on CRAY-2).

The dimensions of the machine are 77" x 30" x 75" with a total footprint of 98 sq ft and weighing 5,000 lb.

Support Environment Low-end machines are air-cooled and used `normal' electricity, but higher performance, more demanding configurations, required special (freon) cooling and power systems. SMARTE (System Maintenance and Remote Test Environment) for remote diagnostics.

Scalability numerous configurations from uniprocessors to a maximum of 8 custom vector CPUs.

1 CPU running in 1987; first deliveries in 1988; nine deliveries in 1989; full production and possible enhancements in 1990.

#### CRAY C90

This architecture has a maximum of 2GB central shared-memory, connected to CPU processors, each consisting of 2 vector pipes. SECDED single-byte correction, double-byte detection memory protection. 64-way vector parallelism with two vector pipes and two functional pipes for each of the 16 possible CPUs.

I/O Subsystem (IOS) with up to 16 I/O clusters, consisting of up to 16 channel adapters per cluster. Optional Solid-state Storage Device (SSD) provides up to 16GB of fast storage for code or swap, with up to four 1800MB/s channels. DD-60 disk drives, with 20MB/s sustained transfer rate to 2GB capacity, and supporting disk striping with 8 disks per IOS/channel.

Performance 16~Gflops claimed peak performance, with 13.6~Gflops achieved in LINPACK massively parallel benchmark.

Data Transfer 13.6GB/s IO bandwidth, and over 250GB/s total memory bandwidth.

One to a maximum of 16 custom ECL vector CPUs.