Today’s biggest supercomputers deliver nearly 40 petaflops, using massive clusters of thousands of servers. But the approach will consumer too much power to scale to systems capable of processing 10 18operations/second.

The US Department of Energy (DoE) and the National Nuclear Security Administration are rallying support for a so-called Exascale Computing Initiative (ECI). The effort aims to deliver by 2023 systems that consume as little as 20 picojoules per operation, 40 times less than today’s computers.

ECI aims to deliver supercomputers that help secure the US nuclear stockpile and seed the high-tech industry with an array of new technologies. But it is not clear where it will get the hefty funding the project will need before its deadline, tentatively set at the end of 2015.

The two agencies have already spent about $200 million on early projects and plan to spend another $91 million next year. Just how much money is needed to hit its ultimate goal is not yet clear, said William Harrod, head of a scientific computing research group at DoE who is championing the program. He hopes to recruit support from other government agencies such as NASA and the National Oceanic and Atmospheric Administration.

“This is the time when people need to come together to create a leap forward,” says Thomas Sterling, a supercomputer expert and professor at the Indiana University. “It’s more than any one industry can do, but there’s a lot of friction between government agencies, and a lot of egos” standing in the way, he told EE Times.

In a recent talk, Bill Colwell, a former program manager at the Defense Advanced Research Projects Agency, called for an exascale project along the lines of the US moon landing or the large Hadron Collider. Such a project would be “incredible in scope, investment and vision… [but] I know it’s possible… and [it] might be required to get to the next stage of computer evolution,” Colwell said.

The US DoE hopes by the end of 2015 to determine how much money it will need to fund an exascale computer project and secure the funds. Click here for larger image

Next page: Intel, Nvidia chip away at exascale