CRI Cray-1A S/N 3

The outlook in 1974 for computational capability beyond NCAR's Control Data Corporation (CDC) 7600 machine was dismal. The CDC STAR, the Texas Instruments ASC, and the ILLIAC IV all seemed unsuitable for the NCAR environment. Two NCAR Computing Facility (CF)* staff members, wondering what had become of Seymour Cray after his departure from CDC and his subsequent 1972 founding of Cray Research, Inc. (CRI), traveled to the CRI Halley Laboratory in Chippewa Falls, Wis., to "see what Seymour was up to."

As CRI's architect, Cray provided the technical vision of a Cray-1 computer with a clock speed that was at least twice as fast as the CDC 7600 and demonstrated balanced scalar and vector performance. The computer was also innovative in its use of reciprocal approximation for division. Although the Cray-1 was still only partially built at the time of the visit, it was apparent that the design reflected a simplicity not found in the other computers. The CF team felt that although many technical and financial objections had to be overcome, the Cray-1 would be a viable upgrade to the CDC 7600.

By 1975, NCAR's 7600 was straining under a severe workload. It was oversubscribed and turnaround time had become intolerable. A request for proposal (RFP) was issued later that year for the acquisition of a “fifth-generation” computer. The RFP and subsequent negotiations resulted in NCAR's order for a CRI Cray-1 system. While the first Cray-1 had been shipped to the Los Alamos National Laboratory in 1976 for a six-month trial period, NCAR was Cray Research's first official customer.

The Cray-1 later became known as the Cray-1A, to differentiate it from the follow-on Cray-1S computer, which had an integrated I/O subsystem. The Cray-1A – serial number 3 – arrived at NCAR on July 11, 1977. The system cost $8.86 million ($7.9 million for the system, plus nearly $1 million for the disks).

The supercomputer weighed 5.5 tons. It arrived in two refrigerated electronic vans, and it took more than 30 construction workers, engineers, and helpers to move it into the computer room. NCAR accepted the Cray-1A in December 1977. It was the first Cray-1A to go into production, and upon its acceptance, Cray Research became a revenue-generating company.

44 yrs after our CRAY-1A #supercomputer was lowered into a computing chamber in the basement of our Mesa Lab, the venerable machine was finally lifted back out. 🏗️



This week, it took a trip north, where it will be displayed in the lobby of the NCAR-Wyoming Supercomputing Center pic.twitter.com/GBIaeEVRwl

— NCAR Science (@NCAR_Science) December 9, 2021

The Cray-1A was very stable compared to the CDC 7600, representing a significant improvement in mean-time-to-failure rates. In good times, the 7600 hardware or software had failed at least once a day – often four to five or more times – whereas the Cray-1A would run reliably for several days and most often failed only because of disk problems.

The CDC 7600 was used as a front end to the Cray-1A. Incoming work flowed through the 7600, which also retrieved archival files needed for use on the Cray1-A from NCAR's TeraBit Memory tape-based mass storage system. Users migrated their jobs to the new system rather slowly at first, and continued to use the 7600 for data-handling chores.

NCAR's Cray-1A had a 12.5-nanosecond clock, eight 64-element vector registers, 1 million 64-bit words (8 megabytes) of high-speed memory and 16 DD-19 high-speed disk drives, each with a capacity of 300 megabytes and a transfer rate of 4.5 megabytes per second. The Cray-1A could chain floating-point vector add and multiply instructions together to produce two floating point operations (FLOPs) per clock; thus the Cray-1A had a peak speed of 160 megaflops. Overall, the throughput of NCAR's Cray-1A was estimated to be about 4.5 times that of its CDC 7600.

When NCAR's first Cray-1A was delivered, CRI was still working on its operating system, Fortran compiler, and other software. Members of Cray's software development staff spent the latter half of 1977 doing development work on site at NCAR. In 1978, Cray's first standard software package was introduced, consisting of the Cray Operating System, the first automatically vectorizing Fortran compiler, the Cray Fortran Translator, and the Cray Assembler Language.

The Cray-1A’s design and vector processing capability were particularly adapted to the needs of the scientific community, and permitted significant advances in the modeling of climate and severe storms. NCAR's first Cray-1A was removed from user production on January 27, 1989, and powered off on February 1 after nearly 12 years of service – the longest production lifetime of any of NCAR's supercomputers.

* The CF, until then a part of NCAR's Atmospheric Technology Division, was reorganized in 1980 as the Scientific Computing Division.