Since the 1960s, leading technology companies have designed ever faster and more powerful supercomputers while chasing the next speed milestone. Until the 1980s, the processing capability of the fastest supercomputers was represented in megaflops, or millions of floating-point operations per second. Later that decade, benchmarks were expressed in gigaflops (billions of flops), and in 1997, the fastest supercomputer achieved teraflop speed (trillions of flops). To match what a 1 teraflop computer system can do in just 1 second, you’d have to perform one calculation every second for more than 31,000 years.
Then the supercomputing industry set its sights on breaking the petaflop barrier — 1 quadrillion, or a thousand trillion calculations per second. Achieving such speeds would demonstrate a nation’s strength in supercomputing and provide a great advantage to whatever scientific, technical and military organizations possessed such power.
In 2008, IBM’s Roadrunner supercomputer became the first to hit the mark. A hybrid system running 12,960 IBM PowerXCell 8i and 6,480 AMD Opteron dual-core processors, Roadrunner achieved a sustained speed of 1.026 quadrillion calculations per second, twice as fast as the next-fastest supercomputer, IBM’s BlueGene/L. To put that into perspective, it would take roughly 6 billion people 46 years of performing one calculation per second on handheld calculators to achieve what Roadrunner could do in a day.
Roadrunner development began in 2002 under the guidance of the project’s chief engineer, Don Grice. Grice joined IBM in 1972 and began working in high-performance computing a handful of years later as part of a team that developed the first IBM general-purpose signal processor.
To build what became the world’s fastest computer, named for the state bird of New Mexico, Grice and the team came up with a first-of-a-kind design. Roadrunner was the first “hybrid” supercomputer, meaning it contained two types of processor architectures — the IBM PowerXCell 8i chip, which was originally developed for video game platforms — and AMD’s x86 processors. AMD’s processors handle the general computational tasks to keep the machine running, while the Cell chips are left to crunch large batches of unstructured data.
Assembled and tested at the IBM facility in Poughkeepsie, New York, Roadrunner comprised enough racks and cabinets to fill a moderately sized warehouse. The system had 80 terabytes of memory. Its 10,000 connections used 57 miles of fiber-optic cable. Twenty-one tractor trailers were required to deliver the system.
Roadrunner broke the petaflop barrier on May 25, 2008, but not without some last-minute drama. The day’s first hurdle was simply to ensure the machine could launch and complete a benchmark test on such a large number of cores, the individual processing units within the computer’s central processing unit. Having passed, the initial attempt to break the barrier suffered a failure in a memory module. In the next attempt, technicians discovered the application they were using wasn’t big enough.
Finally, late that evening, Roadrunner crashed through the petaflop barrier — after most of the team had gone home. Only two technicians were on hand to read the final speed report. “Making the [petaflop] number has caught everyone’s imagination,” Grice said upon reaching the historic milestone. “But it’s really what you can do with that number.”
Despite its size, Roadrunner was remarkably energy efficient for its time, delivering 437 million calculations per watt. That was good enough to land Roadrunner at the third spot on the Green500 list of the most energy-efficient supercomputers. The machine was put to work at the Los Alamos National Lab to ensure the reliability of the US nuclear weapons stockpile and to support national security science projects that require top supercomputer speed and reliability.
Roadrunner’s performance and breakthrough design paved the way for future petaflop supercomputers to facilitate research in many areas of science, industry and society. Petascale machines aid in the design of commercial aircraft and battery systems, conduct comprehensive drug simulations, and create hyperrealistic special effects in movies ranging from Avatar, X-Men and The Lord of the Rings to Black Panther and Dune.
They also aid in the advancement of scientific disciplines, including plasma physics, energy development, astronomy and genomics. In medicine, supercomputers create complex 3D renderings of tissues and bone structures in real time, as patients are being examined. IBM’s Summit supercomputer, unveiled in 2018, has been used to study earthquakes, develop the next generation of materials for energy storage, and investigate supernovae to gain a better understanding of the origins of the universe.
High-performance supercomputers can also analyze cause and effect in capital markets and solve complex physics problems. General Electric is using IBM's Summit in a project with the Department of Energy to boost the production of cleaner power.
Beyond petaflops are exaflops, processing speeds marking a million trillion calculations per second. That’s 1,000 times faster than petaflop-class systems. IBM was one of six technology providers to be awarded a contract in the Department of Energy’s Exascale Computing Project. The company’s first exascale supercomputer, Aurora, is scheduled for deployment at the Argonne National Laboratory.
Exascale machines hold the potential to hasten the pace of discovery in highly complex areas that have long been beyond the reach of computers — from hurricane prediction and long-term climate modeling to tools development for fusion energy, medicine and materials. Aurora is intended to focus on investigating low-carbon technologies, subatomic particle behavior, cancer therapies and more-efficient solar cells.
When it comes to supercomputers, there’s always another barrier to break. And IBM continues to be at the vanguard.
It ushered in an era of high-performance computing that valued efficiency as much as raw processing power
The world’s most powerful mainframes continue to push technological boundaries
With ‘Stretch,’ IBM entered the age of supercomputing