We are heading into the Supercomputing 2011 (SC’11) conference next week. The high performance community is driving hard towards building extreme scale supercomputers with the next big milestone set at building an exaflop supercomputer – that’s roughly the performance of 70 million laptops put together. These “exascale” supercomputers are essential to drive innovation in science and technology.
In a freshly published report, industry analyst firm IDC argues that the fastest path to this exaflop milestone is through heterogeneous designs. They state that x86 processors will not be enough to meet the performance and power goals that the U.S. Department of Energy (DOE) has outlined – 1,000 times faster than the current petaflop systems and 50-100 times more energy efficient. Given the daunting challenges of cost, power consumption, application performance and space, IDC concludes that heterogeneous systems offer the best solution to achieve exascale performance.
IDC analysts Earl Joseph and Steve Conway write, “IDC believes that heterogeneous computing will be indispensable for achieving exascale computing in this decade.”
Their report, entitled “Heterogeneous Computing: A New Paradigm for the Exascale Era,” is available here as a PDF document.
The DOE has a stated goal to achieve exascale performance by the end of the decade, but the U.S. isn’t alone in its desire to reach this threshold. China, Russia, Japan and the European Union are all racing toward the same goal, because of the tremendous scientific, industrial and economic benefits these extreme scale supercomputers will provide. We’re not just talking about theoretical science. Imagine the economic advantages accruing to the country that develops safe, non-polluting fusion reactors, in terms of prosperity, national security and job opportunities.
Exascale systems, the DOE believes, will enable it to research cars that use less fuel, develop clean and abundant fusion power, mitigate global warming and even develop cures for diseases like cancer and diabetes.
And like the space race of the 1960’s, there will be hundreds or thousands of technological dividends that filter down from such research. And just like the space race, there is a race on, to get to the next level of supercomputing.
Oak Ridge National Laboratories (ORNL) has thrown down the first gauntlet in this exascale race by announcing their intention to build a 20 petaflop heterogeneous supercomputer accelerated by NVIDIA Tesla GPUs.