Jeff Nichols, associate lab director in charge of scientific computing at Oak Ridge National Laboratory, visited China for the first time earlier this month and spoke at an Nvidia-sponsored conference in Beijing.
The folks at the Beijng conference included representatives from the top scientific computing labs, including those housing the No. 1 (Japan’s K Computer) and No. 2 (China’s Tianhe-1A system) on the latest list of the world’s Top500 supercomputers. Nichols visited the nearby lab housing the Tianhe-1A computer as part of the conference, where he spoke about the scientific computing at ORNL — including the transformation of the No. 3-ranked Jaguar.
Nichols said it’s going to be difficult for the United States to gain or reclaim world leadership in supercomputing.
China is a particularly formidable competitor, he said. “They’ve got a lot of money. They can invest tons of money in computers. They’re going to be hard to compete with. I think we have a better handle on applications and scalability and issues surrounding scalability, especially these unique heterogeneous nodes . . . But they can afford to build and buy these machines a lot faster than we are.”
In the next three to four years, China will graduate students with a lot of experience on these advanced machines and the U.S. could lose headway, Nichols said.
“It’s a race. It’s hard to say,” he said. “You’ve got people like China and Japan who are willing to put a billion dollars into (a supercomputer). When money is no exception, it’s tough for us to compete. We can innovate and understand applications and be smarter and develop applications in working with vendors. Whether that’s enough to retain (world leadership), it’s a crap shoot.”
Who’ll be first to achieve exascale computing?
“The problem is when we say exascale we’ve got to go through at least two more generations,” Nichols said. “We may have this same conversation a couple of times.”
ORNL, which is home to perhaps the world’s best collection of scientific supercomputers, is in the process of transforming the current Cray XT5 Jaguar (capable of 2.3 million billion calculations per second and ranked No. 3 in the world) into a Cray XK6 system that will become a 10-20 petaflops machine in the next year or so and perhaps even greater capabilities beyond that.