50 years of the Cray-1: the supercomputer that looked like a sofa
Fifty years ago, the Cray-1 set new standards in computer technology. Its computing power, unmatched at the time, made it a symbol for an entire generation of supercomputers.
At first glance, the Cray-1 could be straight out of a science fiction movie. Twelve tall cabinets form an open circle, surrounded by an upholstered bench. The strange design had a clear purpose: it was meant to make the computer as fast as possible. And it worked. From 1976 to 1982, the Cray-1 was considered the world’s fastest supercomputer.
Why every inch counted in the Cray-1
The computer was developed by American engineer Seymour Cray. In designing the system, he had a clear goal in mind. The electronic components should communicate with each other as quickly as possible.
To achieve this, he kept the connections between every component as short as possible. After all, the longer a wire, the more time it takes for an electrical signal to travel through it. And given the Cray-1’s high clock speeds, every inch counted. The C-shaped layout cut down on distances between the components.
Source: Clemens Pfeiffer/Wikimedia Commons/CC BY 2.5
In total, there were more than 96 kilometres of wire criss-crossing the machine. Nevertheless, no single link was longer than about 90 centimetres. Most of the wires were laid by hand. It took nearly a year just to assemble a Cray-1.
The padded bench also had a practical function: parts of the power supply were hidden underneath it. And since the tightly packed electronics also generated a lot of heat, the Cray-1 used a special cooling system with a liquid refrigerant to dissipate it.
Source: Image: Rama/Wikimedia Commons/CC BY-SA 2.0 FR
The result was a computer around 1.95 metres tall and weighing more than five metric tons. Its unique shape earned it the nickname, «world’s most expensive loveseat.» A justified moniker, since several people could sit on its upholstered bench.
160 million arithmetic operations per second
The inner workings of the Cray-1 were also extraordinary for their time. Its processor ran at 80 megahertz and achieved a theoretical peak performance of 160 megaflops. That’s equivalent to 160 million floating-point operations per second. Calculations involving decimal numbers like these are important in scientific simulations, for example.
The number sounds pretty modest today. Modern processors and graphics cards can perform many trillions of such operations per second when handling suitable tasks. Even older personal computers significantly exceed the peak performance of a Cray-1.
Its strength lay primarily in what’s known as vector processing. The computer was able to apply the same arithmetic operation to a whole series of numbers. For example, processing thousands of measured values one after another without needing a new command for each value. This sped up calculations involving large datasets in particular and made it efficient for scientific tasks.
Source: The Board of Trustees of the Science Museum
The massive performance leap it provided soon became apparent at the National Center for Atmospheric Research (NCAR). There, the Cray-1 achieved a computing throughput approximately 4.5 times higher than that of the CDC 7600, which had been used previously.
In a typical configuration, the RAM totalled about eight megabytes. While barely enough for a single high-resolution photo today, it was a considerable amount of storage back then.
However, this performance came at a price. Depending on its configuration, a Cray-1 cost several million US dollars. For example, NCAR paid 8.86 million dollars for its system in 1976. The hard drives alone accounted for just under one million dollars.
Operating costs were just as high. The computer alone required about 115 kilowatts of electrical power. In addition, the complex cooling system and regular maintenance resulted in high costs.
A supercomputer for science and the military
Before developing the Cray-1, Seymour Cray had already worked on high-performance computers at the Control Data Corporation. Then, in 1972, he founded his own company, Cray Research, in the state of Wisconsin.
Source: Cray Research
Its most important clients included government research institutions. During the Cold War, the USA invested massive sums into the development of high-performance computers. Nuclear weapons research, in particular, required complex physical simulations.
The Los Alamos Scientific Laboratory (LASL) in New Mexico received unit number one in March 1976. That facility played a central role in the development of the atomic bomb during World War II. After a six-month trial period, the LASL put its new computer into regular operation in October 1976.
Climate and weather research also benefited from its computing power. Starting in 1977, NCAR began using its own unit to run models of the atmosphere and severe storms, among other things.
For Cray Research, the machine proved to be a success. A total of about 80 units were sold. Thanks to that, the company established itself as one of the leading manufacturers of supercomputers. The system at NCAR remained in operation until February 1989.
50 years later: a Cray-1 made from old Macs
A recent project out of Spain illustrates just how much computer technology has advanced since then. To mark the anniversary, the «Museo de Historia de la Computación» has built a full-scale replica of a Cray-1. The replica took about 18 months to build.
Source: Museo de Historia de la Computación
From the outside, the replica is modelled after the historical original. Inside, however, there are 30 decommissioned Mac Minis from 2012 to 2014. Together, their 70 processor cores distribute computations among themselves. In a test with 64 active cores, the cluster achieved approximately 1.3 teraflops – 1.3 trillion floating-point operations per second. To do this, the museum had the computers multiply large matrices of numbers. The main purpose of this recreation is to demonstrate how multiple processors work together to perform computations.
However, the results can’t be directly compared. The original’s figure represents theoretical peak performance, while the replica’s result stems from a specially developed computational test.
The museum’s already planning an expansion that’ll include up to 80 modern Mac Minis equipped with Apple’s M4 chip. Visitors can see how calculations are distributed across multiple processors by observing the simulation. In other words, the project covers both technological progress as well as the evolution of supercomputer architecture. While the original Cray-1 relied on a particularly fast vector processor, the replica distributes its computational tasks among numerous independent computers.
My interests are varied, I just like to enjoy life. Always on the lookout for news about darts, gaming, films and series.
Interesting facts about products, behind-the-scenes looks at manufacturers and deep-dives on interesting people.
Show all
