Seymour Cray was an electrical engineer and visionary supercomputer architect who spearheaded a series of computers deemed the fastest in the world for decades. Cray’s design philosophy hinged on two key principles: eliminate heat and make sure that all signals intended to reach a destination at the same time do indeed arrive at the same time.
Seymour Cray designed a series of supercomputers that claimed the title of fastest in the world for decades.
His legacy continues today with the latest generation of supercomputers running on over 50,000 processors.
An Innovator at Heart
As a young boy, Cray harbored his interests in science and engineering in the basement of his family home, which he converted into a “laboratory.” There, Cray was rumored to have built a device from Erector Set components that could turn punched paper tape into Morse code signals.
Once Cray graduated from high school, he was drafted for World War II as a radio operator, where he was assigned to break Japanese naval codes. He returned home to earn a B.Sc. in electrical engineering at the University of Minnesota and a M.Sc. in applied mathematics.
Even after Cray started his career, he preferred the research lab to the office. At his first job at Engineering Research Associates (ERA), Cray developed a magnetic drum for the U.S. Navy, a component for the working memory of the first computers. His most notable achievement at the ERA was designing the ERA 1103, a centralized vacuum tube computer system that Cray programmed by hand with paper tape.
ERA’s ad for magnetic storage systems. Image used courtesy of the Engineering Research Associates via Computer History
Although Cray worked with digital technology for most of his life, he maintained a somewhat analog mindset that allowed him to view computer systems holistically. He was known for his simple design approach and leveraged traditional tools and systems long after he had advanced them.
The ‘Cray Way’ of Transistorised Computers
Due to corporate restructuring at ERA, Cray left to found a new company, the Control Data Corporation (CDC), where he created the CDC 1604 and CDC 6600.
The CDC 1604 was the first transistorized computer that overcame overheating vacuum tubes—a key performance flaw of early computers—and embraced solid-state devices on a grand scale.
CDC 6600 system console. Image used courtesy of Jitze Couperus via Wikimedia Commons (CC BY 2.0)
“Anyone can build a fast CPU. The trick is to build a fast system,” Cray once remarked. With this mindset, Cray also focused on maximizing I/O bandwidth to avoid “starving” the processor of data to compute. When it was released in 1964, the CDC 6600 was the fastest computer in the world. Cray replaced I/O interrupts with polled request issues by a peripheral processor, which made all the transfers to and from the CDC 6600’s central memory. This architecture solved the issue of imprecise interrupts—a problem that confounded competitors like IBM.
The CDC 6600’s architecture also prompted the adoption of floating-point arithmetic operations as a standard for measuring computer performance. The computer featured three million floating-point operations per second (FLOPS), Freon cooling, and innovative packaging that led to huge commercial success, with 100 units sold in the following years.
The 6600’s successor, the CDC 6700, further boosted the 6600’s peak computational speeds tenfold.
Cray-1: The First Supercomputer With Vector Processing
In 1972, Cray founded his own research company, Cray Research Inc., where he designed the renowned Cray-1. The first Cray supercomputer, released in 1976, used a vector processing architecture that could perform 240 million calculations per second and had a main memory capacity of up to 8 megawords or 64 megabytes. In designing the Cray-1, Cray once again underscored the importance of ensuring the entire computer’s speed, not just the processor.
Cray with the Cray-1 supercomputer. Image used courtesy of the Star Tribune
Cray-1 was built cylindrically with components arranged around a central structure. To maintain the system’s optimal temperature, it contained a circulating liquid cooling system based on Fluorinert (C10HF22N). The supercomputer beat all other computing systems on the market by a wide margin. Los Alamos National Laboratory borrowed Cray-1 in 1976, and the National Center for Atmospheric Research (NCAR) bought the first commercial version of the system for $8.8 million.
The Cray-2 Falters—Yet Cray Persists
Cray’s work wasn’t always perfect. The next version of his supercomputer, Cray-2, created in 1985 after he resigned as CEO of Cray Research, wasn’t as commercially successful as its predecessor. The Cray-2 competed with a more successful model, the two-processor Cray X-MP, built by a different team. The Cray-2 was also released toward the end of the Cold War when the company had less money to spend on supersized computers priced in the millions per piece.
Cray didn’t let design setbacks or budget issues defeat him, however. He began working as an independent contractor for Cray Research and later founded another company called SRC Computers. His goal at SRC Computers was to design his own massively parallel machine—a notion he had previously resisted early in his career in favor of a single fast processor.
“If you were plowing a field, which would you rather use: two strong oxen or 1,024 chickens?” he had famously asked as a young engineer. This logic fell through by the mid-1980s, however, with the rise of modern compiler technologies, and Cray jumped onboard to solve the bottlenecks plaguing communications and memory performance in many parallel designs. His plans were sadly cut short by his sudden death in a car accident in 1989.
Cray’s Supercomputing Philosophy and Legacy Lives On
Cray achieved his goal of building the fastest computer in the world, in part because his computers included a built-in cooling system. This system extended into the mainframes and thermally coupled to metal plates in the circuit boards and other coolant-immersed systems.
His computers were also remarkable because every signal path was the same electrical length, solving the issue of skew and ensuring that values to be acted upon at a certain time were all valid values. To achieve this uniformity, Cray was known to run traces back and forth on a circuit board until he achieved the right length. He would then run Maxwell’s questions to account for any radio frequency effects that may have altered the signal velocity and electrical path length.
Hewlett-Packard continues Seymour Cray’s legacy with a series of advanced Cray computers. After decades of improving the Cray-1, updated models of the supercomputer took off with the rise of Big Data. The current world supercomputing champion, the Hewlett Packard Enterprise Frontier, can calculate more than one quintillion (1,018) FLOPS and 1.5 exaflops at peak performance.
The supercomputer Frontier at Oak Ridge National Laboratory. Image used courtesy of MIT Technology Review
Cray is today regarded as the father of supercomputing and even “the Thomas Edison of the supercomputing industry.” The IEEE established an award in Cray’s name, the IEEE Computer Society’s Seymour Cray Computer Engineering Award, which celebrates contributions to high-performance computing systems that embody Cray’s engineering ingenuity.
