News

China Threatens US Supercomputing Supremacy

• By Henry Kenyon
• 06/03/2010

According to the New York Times, the Dawning Nebulae, housed at the National Supercomputing Center in Shenzen, China, has reached a sustained computing speed of 1.27 petaflops. A petaflop is one thousand trillion mathematical operations per second. That performance puts it right behind the Energy Department's Jaguar supercomputer. Located at the Oak Ridge National Laboratory, Jaguar has a speed of 1.75 petaflops and a theoretical peak capacity of 2.3 petaflops.

The ranking announcement was made on Monday at the International Supercomputer Conference in Hamburg, Germany. Presented twice a year, the TOP500 list ranks the world's 500 fastest computers. According to the Top500 release, the Chinese computer is ranked fastest in pure theoretical performance, with a top processing speed of nearly 3 petaflops, but this is considered less important than the actual computing speed reached on a standard computing test.

Chinese designers achieved the high speeds by using a mix of commercial processors. The supercomputer consists of a Dawning TC3600 Blade system with Intel Xeon 5650 processors combined with specialized NVidia Tesla C2050 graphics accelerators.

Computer scientists speculate that the Dawning Nebulae supercomputer is part of an ongoing effort by China to produce the world's fastest computing platform. Another Chinese computer, made from Chinese-designed and manufactured processors and components, is scheduled to be unveiled later this year. China has been steadily moving up the computer speed ladder over the last several years. In the previous year's TOP500 list, a Chinese supercomputer ranked fifth in overall speed.

The Chinese computer progress is a natural outgrowth of the nation's ongoing development, said Buddy Bland, project director for the Oak Ridge National Laboratory's Leadership Computing Facility. "I think the Chinese have a lot of money to invest and they're looking to modernize their society. They understand that high-performance computing is a tool that can help with economic growth," he says.

Bland said the Chinese are interested in using high-performance computing to help in areas such as manufacturing and product design, and energy-related research in areas such as oil field recovery.

As China is developing its next-generation petaflop supercomputers, the U.S. government and international partners are laying the groundwork for the next generation of high-performance machines. The proposed Exascale Computing Initiative will potentially involve several U.S. government agencies such as DOE, which oversees Oak Ridge. However, Bland cautioned that the initiative has not been approved by the department's senior management. "At this point it's a proposed initiative, not an ongoing one where there is funding coming from the federal government to support it," he said.

Despite the lack of an official green light for funding, there is a great deal of interest in developing exascale machines — computers that are a thousand times faster than petaflop computers. Bland said the effort's goal is to have an operational exascale computer by the end of the decade. "We are working very hard to understand how to get there," he said.

A decade ago, when petaflop supercomputers became possible, there was a relatively clear development path from the teraflop computers then in use. "We had a reasonable idea of how we were going to program these computers and what the systems might look like," he admits.

But exascale computers present many new challenges. Bland said that computer scientists must overcome issues such as programming models, new hardware, memory and file systems, application needs and fault tolerance. The program will also need substantial investment, research and development to make headway on those multiple fronts. "Today we know more about what they [exascale computers] won't look like than what they will look like," he says.

The scale of the new effort will require not just a national, but a global effort. "We think we're going to need everybody to be able to work together to be able to get to the exascale systems in this decade," he says.