UHPC Developments Move DARPA Closer to Goals
Richard Murphy, a computer architect at Sandia National Laboratory, recently weighed in on progress toward the goals set forth by the Ubiquitous High Performance Computing program (UHPC). For those who are not familiar, this initiative, which was set forth by the Defense Advanced Research Projects Agency (DARPA) aims to bring petascale and exascale computing innovations into military use via a program of focused research efforts on everything from power and efficiency to performance to applications.
The program, which got its start last year posed a challenge to scientists to build a petaflop system that consumes no more than 57 kilowatts of electricity, in part so that the military could bring computing power out of large datacenters and into the field for immediate, on-spot use. Aside from this more practical military use of high-end HPC systems on the fly, massive benefits for computing efficiency for cost savings and reduced environmental impact would be realized as well.
To bring the kilowatt usage down to the challenge level of 57 kilowatts is no simple task; it will require a dramatic, almost unthinkable reduction in electricity use—all the while retaining the key performance required for military high performance computing applications.
Teams working on such initiatives are vying for the chance to win an award to build a supercomputer for DARPA. Those who come close to the power goals will need to dramatically rethink how computers are designed, particularly in terms of how memory and processors move data. As Discover Magazine pointed out, “The energy required for this exchange is manageable when the task is small—a processor needs to fetch less data from memory. Supercomputers, however, power through much larger volumes of data—for example, while modeling a merger of two black holes—and their energy can become overwhelming.”
According to Richard Murphy, “it’s all about data movement.” Those in the race to meet DARPA’s challenge are seeking ways to make data movement more efficient via distributed architectures, which clip the distance data travels by the addition of adding memory chips to processors. “We move the work to the data rather than move the data to where the computing happens,” Murphy says.
As Eric Smalley wrote today following a discussion with Richard Murphy:
“Sandia National Laboratory’s effort, dubbed X-caliber, will attempt to further limit data shuffling with something called smart memory, a form of data storage with rudimentary processing capabilities. Performing simple calculations without moving data out of memory consumes an order of magnitude less energy than today’s supercomputers.”
Full story at Discover Magazine