The US Department of Energy (DoE) has big plans for the application of supercomputing technology, and at the moment, exascale computing represents the ultimate goal. Achieving this thousandfold increase over existing petascale computing performance is no easy task, so the DoE has partnered with the National Nuclear Security Administration (NNSA) in an effort to make it a reality by 2020.
In order begin the research and development process, the FastForward program was created to align leading technology companies, such as Intel, AMD and NVIDIA, with the government’s national labs in the pursuit of new processor, memory and storage technology.
IBM’s Sequoia was recently crowned as the world’s fastest supercomputer. Installed inside Lawrence Livermore National Laboratory, Sequoia’s LINPACK performance of 16.32 petaflops, represents a 55% gain over Fujitsu’s 10.51 petaflop K Computer, the previous supercomputing champion, but even that level of performance is a far cry from the exascale objective at hand.
FastForward represents the initial phase of the Extreme-Scale Computing Research and Development program as it will take many years to develop the advanced technology required to achieve exascale performance. Beyond the raw computing requirements, one of the most daunting challenges involves powering an exascale computer. At current levels of power consumption, something on the order of a 1,000 megawatt power plant will be required, and that equates to the output of a nuclear reactor.
Shifting many of the computational tasks from CPUs to GPGPUs may hold the key to reducing this power requirement. According to Bill Dally, NVIDIA’s Chief Scientist, the GPUs in an exascale system built with NVIDIA Kepler K20 processors would consume just 150 megawatts of power, a dramatic decrease.
National Lab Consortium
- Argonne National Laboratory
- Lawrence Berkeley National Laboratory
- Lawrence Livermore National Laboratory
- Los Alamos National Laboratory
- Oak Ridge National Laboratory
- Pacific Northwest National Laboratory
- Sandia National Laboratory
What’s interesting to note is that all of the technology developed along the way will ultimately be used outside the laboratory and become incorporated into commercial applications. This idea is spelled out in the FastForward statement of work.
“While DOE’s extreme-scale computer requirements are a driving factor, these projects must also exhibit the potential for technology adoption by broader segments of the market outside of DOE supercomputer installations.”
For example, Intel will be incorporating technology from their Many Integrated Core (MIC) architecture, which will also find use within commercial applications that involve Big Data. According to Mark Seager, CTO for Intel’s High Performance Computing Ecosystem group, they will need to employ some “radical approaches” in order to reduce system power requirements by a factor of 2 to 3 times, which may be achievable by leveraging aspects of their previous near-threshold voltage circuitry research.
FastForward Subcontract Awards
- Intel: $19 million for both processor and memory technologies
- AMD: $12.6 million for processor and memory technologies
- NVIDIA: $12.4 million for processor technology
- Whamcloud: $8 million for storage and file system technologies