To get an edge over China in the supercomputing arms race, the U.S. plans to build a 180-petaflop supercomputer that will be used mainly for scientific research.
The U.S. Department of Energy supercomputer, called Aurora, will be deployed by 2018. Its peak performance of 180 petaflops is more than three times the performance of the fastest supercomputer today. But other systems are already being built to deliver between 150 petaflops to 300 petaflops of performance.
Supercomputers are vital to economic, security, scientific and environmental programs in the U.S. and other countries. For example, supercomputers are used for military programs, weather forecasting and economic model simulations. Countries also boast about their supercomputers as a sign of technological superiority.
The world's fastest supercomputer today is China's Tianhe-2, which can deliver peak performance of 54.9 petaflops, according to a compilation by Top500.org. The Aurora system will be an upgrade over the DOE's Titan supercomputer, the world's second fastest.
Aurora could be one of the fastest supercomputers by 2018, if not the fastest, said Dean McCarron, principal analyst at Mercury Research.
New supercomputers also provide a gateway for new processor, memory, networking and storage technologies. Many technologies typically trickle down from supercomputers into servers used in data centers.
The Aurora supercomputer will be deployed at the Argonne National Laboratory in Lemont, Illinois. The DOE is investing US$200 million in the supercomputer, and contracts were awarded to Intel and Cray to integrate the system.
Aurora will be based on a design of Cray's next-generation supercomputer code-named Shasta. Not much is known about Shasta yet, but it'll have a faster interconnect than Cray currently provides for supercomputers. A high-bandwidth network is key in reaching the 180 petaflops of performance, as it helps in faster data transfers between storage, processors and memory units inside a supercomputer.
Supercomputers are giant multi-processor systems. Intel said the computer will combine "next-generation" Intel Xeon -- which are used in multi-socket servers -- and Xeon Phi chips, which have up to 72 cores. Intel is also planning on using silicon photonics technology, with data transport inside supercomputers happening through pulses of light.