Cray-2

Seymour Cray began to create a new supercomputer - Cray-2 - even before the completion of the refinement of the previous model - Cray-1 , that is, approximately in 1976. In the initial version, Cray-2 was a machine with four vector processors with a cycle time of 4 ns, that is, three times less than that of Cray-1 ( 12.5 ns ). Until 1978, Cray-2 was only in projects, since Cray and his team were busy assembling the production Cray-1 and their improvements according to customer requirements. In early 1978, Cray decided that Cray-2 would not contain 4 vector processors, but 64 scalar processors. It was a radical departure from the successful direction of vector processors to the little-known technology of mass-parallel computing. For such an unusual computer, a special parallelizing compiler was required, which Steve Nelson took up work on. After several months of work, Cray paid a visit to the Livermore National Laboratory , where he introduced a new future machine to the programmers there. It turned out that the 64-processor machine is not in demand, and Cray had to curtail work on ahead of his time technology and again return to the original circuit.

To increase the clock frequency, the distance between the elements of the computer had to be reduced. If the longest wire in Cray-1 was no longer than 6 feet, in Cray-2 the wire length was reduced to 16 inches. Tighter packaging meant greater difficulty in heat dissipation. Attempting to use the Cray-1 approach has failed. The distance between the panels was so small that between them it was impossible to place the copper plates of the heat sinks for cooling. The Cray-2 project began to resemble the failed CDC 8600 project, which is why Cray took a bold step. He proposed setting up a laboratory in Boulder , Colorado, which would be involved in the creation of integrated circuits. In Cray-1 and initially in Cray-2, the simplest microcircuits of a small degree of integration were used with only two logic gates on the emitter-coupled logic of bipolar transistors . The laboratory in Boulder was supposed to design and create ultra-large integrated circuits for Cray-2, which would allow to pack a large amount of logic into the small dimensions of the machine. The city of Boulder was not chosen by chance, since there was the headquarters of the - the first buyer of Cray-1 and a potential buyer of Cray-2. The lab, called “Cray Labs,” was designed to place the entire Cray-1 computer on one or more chips, for which the lab staff took one copy of Cray-1 out of the cogs to understand how it works. At the same time, work on Cray-2 at Cray Research was frozen, and team engineers were transferred to other areas.

In 1981, Seymour Cray introduced a radical design proposal for the Cray-2. He suggested placing the entire computer in an inert liquid called Fluorinert from 3M , which was used in medicine at the time for heart surgery. The fluid had to circulate around and inside the computer and remove heat more efficiently than air. This breakthrough made the laboratory in Boulder work on special VLSIs unnecessary, and in 1982 it was closed. In 1980, Cray resigned as head of Cray Research and began working in the company as an independent contractor. This allowed him to devote more time to Cray-2, work on which lasted for 4 years.

Cray-2 was introduced to the public in 1985. With a peak performance of 1.9 GFlops , it became the fastest supercomputer in the world, having displaced the Cray X-MP , released in 1983, from the pedestal. This title was held by Cray-2 until 1990, when ETA (a division of CDC ) released the ETA-10G supercomputer.

• Processor: 2 or 4 vector processors
• Element base: IC with 16 logic gates , emitter-coupled logic on bipolar transistors
• Tact Time: 4.1 ns
• Memory Capacity: 256 million 64-bit words
• Cooling: liquid , fully immersed
• Operating System: UNIX-like Unicos or Cray Operating System
• Software: two Fortran language compilers : CFT2 and CFT77 with automatic code vectorization , C language compiler, CAL macro assembler, utilities and libraries for working with input-output devices and organization of task execution
• Power Consumption: 195 kW
• Dimensions: height - 114.3 cm , diameter - 134.6 cm , 14 vertical pillar columns assembled in an arc 300 degrees long

In 1985, the cost of Cray-2 was 17.6 million US dollars ^[1] .

In the summer of 1985, the first copy was delivered to the Livermore National Laboratory , where the Cray-2 prototype had long been working. Initially, Cray-2 was developed for the Department of Defense and the US Department of Energy . It was supposed to be used for research in the field of nuclear weapons and oceanography . However, Cray-2 was used for peaceful purposes, for example, in NASA (the 2nd serial number was delivered to the Ames Research Center in September 1985 ^[2] ), universities and corporations around the world.

Simultaneously with the development of Cray-2, Cray Research created the Cray X-MP machine, and Japanese counterparts of the Cray-1 from NEC , Fujitsu and Hitachi appeared on the supercomputer market. In order to make Cray-2 more attractive to the customer, its memory system was significantly redesigned, both in terms of its volume and in terms of its speed. When the machine was finally introduced to the market, it was already so far behind modernity that all its performance advantages were manifested mainly due to this fast and large memory. Buying Cray-2 made sense only for those organizations that required fast processing of large amounts of data. Thanks to the properties of Cray-2 memory, computer simulation was able to move from two-dimensional models and approximate three-dimensional to accurate three-dimensional models.

In 1987, Cray-2, owned by the British Atomic Energy Commission, was used to study the spread of fire at London's Kings Cross Station ^[3] .

On February 19, 1992, using the Cray-2 supercomputer, mathematicians David Slowinski and Paule Gauge found the largest prime number M ₇₅₆₈₃₉ at that time . 10,000 hours were spent searching for it and another 19 hours of machine time were needed to prove that the number is prime ^{[4] [5]} .

In 1990, specifically for the Livermore National Laboratory, a single-copy 8-processor Cray-2 was built worth $ 19 million ^[6] . By then, Seymour Cray had separated from Cray Research and created the new company Cray Computer Corp. to work on the new Cray-3 computer and support for Cray-2 (Cray Research focused on the development of the Cray X / Y-MP line and the new development - Cray C90 ). The Livermore Laboratory was the first customer of the still non-existent Cray-3 , and Cray Computer Corp. to meet customer expectations. built this 8-processor model.

Cray-3 was supposed to replace Cray- 3 , however, due to problems encountered during development, only one instance of Cray-3 was built. Some ideas from Cray-2 were developed in the Cray X1 computer.

•  

  Heavy Duty Logic Boards

•  

  Modular rack close up

•  

  Cray-2 at the Museum of Arts and Crafts in Paris

•  

  The only 8-processor Cray-2 (serial number 2101) at the Museum of Computer History in the US, California

• Cray Inc. - the successor company to Seymour Cray and the successor to its development

• Charles J. Murray. The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer . - Wiley, 1997 .-- 232 p. - ISBN 9780471048855 . (English) - a book about ERA, CDC, Cray Research and a detailed biography of Seymour Cray

• Photos of Cray-2 modules
• Cray-2 Advertising Brochure
• Cray-2 Documentation at bitsavers.org: Technical Description ; Functional Description