The Steve Jobs of supercomputers: We remember Seymour Cray

Fast, cool, simple. Repeat

Startup king strikes

The catalyst for all of this was Seymour Cray, who graduated with a degree in electrical engineering and a masters in applied mathematics before he began work on building cryptographic equipment for the US Navy. It was there where he designed his first computer – the 1103.

Cray left the Navy to form the first of four companies in 1957 – Control Data Corp. Here he built the world’s first fully transistor-based computer, the CDC 1604.

What came next is regarded as the world’s first super: the CDC 6600. The 6600 delivered nine million floating points per second (MFLOPs) – making it the fastest computer of the time, faster than IBM’s 7094 which ran at 100 kiloflops (Kflops) and was used to process scientific workloads.

The advent of the CDC 6600 reputedly infuriated IBM’s chairman Thomas J Watson. History says he fired a memo at staff demanding to know how a firm with just 34 staff – including the janitor – could have beaten his mighty IBM.

The big difference was that Cray's sole interest was the high-end scientific market, driven by high speed and data flow. Speed and high performance were therefore baked in as standard. IBM's overall focus was diluted by the need to run ordinary business opeations, such as payroll, with its machines. Cray believed speed and data flow were best served by simplicity – a simple architecture and a simple design philosophy.

Dag Spicer, senior curator of the Computer History Museum in California, told The Reg: “He used very conservative design techniques that contributed to the success of the systems – he made circuits very tolerant of noise and environment, so they were rock solid.”

Bill Sembrat, who worked with Cray, reckoned Seymour's great advantage was his ability to see the "whole picture", yet pick out and concentrate on the smallest details.

“He had great ability to get the heart of the problem,” Sembrat said. "Many solutions were very simple – after seeing the solution you were surprised at how simple it seemed and would come to the conclusion: 'It was so simple, why didn't I see that?'"

Sembrat added:

He worked very hard on making the world's fastest computers. His aim was always to make a computer with few numbers of processors and really concentrate on one processor. He always [worked] through curves when asked about systems with many processors, saying thing things like: ‘There are enough people working on these systems’ meaning I'm not interested. We did have a system at Cray Computer whose goal was 30 million processors. Naturally the processors were very small and the goal was that they had to be very, very cheap.

Speed came through architectural principles and concepts which would become the hallmark of the CDC 6600 and future Cray supers. He employed what would later become known as Reduced Instruction Set (RISC), which would be seized upon later by Sun Microsystems and IBM.

His systems were compact on the macro and micro level. The idea was to cut the size of the comms networks while Cray employed materials that were fast and dense. The CDC 6600’s transistors were made of germanium, which switched in less than three nanoseconds – silicon transistors were too slow for early computers. The CDC 6600 packed 600,000 transistors on to a single cordwood module to reduce the lengths of the connecting wires.

"He worked very hard on making the world's fastest computers. His aim was always to make a computer with few numbers of processors and really concentrate on one processor" - Bill Sembrat

Speed and density produced the kind of heat that defeated conventional air cooling, so Cray employed Freon – halocarbons used to chill fridges and freezers. Cooling would be come a constant feature and he went on to lodge a number of patents in this field.

Sembrat told us: “I consider the 6600 one of his greatest accomplishments.” He might be right, as the CDC 6600 encapsulated the ideas that flowed in Cray’s later systems. He left CDC in 1971 to form Cray Research, which delivered the Cray-1 in 1976 and effectively began one man’s dominance of the super market.

Again, speed was paramount, with a reduced footprint helping: a single-core processor was nudged along by an architecture that saw no connecting wire longer than four feet. To help, the Cray-1 employed a unique and very 1970s looking feature - a C-shaped tower. Again, heat was a challenge, as he’d used low-density/very-high-speed emitter-coupled logic circuits, rather than high-density/slower-speed devices. These circuits generated plenty of heat in their tight cabinet.

The Cray-1 used Freon for cooling, although this was replaced by 3M's Fluorinert in the successor Cray-2. The super's components were immersed in hundreds of gallons of the stuff. The result was 160 megaflops (MFLOPs) worth of processing in the Cray-1, which came with 8MB of memory and was priced at $8.8m at the time, or $34m adjusted to today.