The race is on to design machines that can keep pace with big data.
The Energy Department, in a cost-savings gambit, has entered into a deal with several technology companies to prototype extreme-scale supercomputers the government needs by 2020 for national security but cannot afford right now.
The “FastForward” research and development consortium, comprising Energy national laboratories and high-performance computing firms, expects to support the production of economical exascale systems by the end of the decade.
As of last week, the department had divvied up at least $43.6 million for processor, memory and storage designs. Intel, Nvidia, AMD and Whamcloud have announced respective awards of $19 million, $12 million, $12.6 million and an undisclosed amount.
Exascale machines will be able to perform a quintillion -- or a billion billion -- floating point calculations, “flops,” per second, meaning they will be 1,000 times faster than a one petaflop supercomputer. A 16 petaflop Lawrence Livermore National Laboratory system currently is the fastest computer in the world.
Energy will kick in about $60 million to accelerate the worldwide race to build an exascale machine, according to a March 29 solicitation for grant proposals. It is not clear if other firms will receive funding because the government has not publicly announced the awards yet.
Intel Federal President David Patterson said during an interview, “our vision for supercomputing is to really enable the current and future generation of scientists to achieve breakthroughs that they couldn’t otherwise with the problems that they face.” Intel acquired fellow awardee Whamcloud on Friday.
Whamcloud officials said in a statement, “the developed technology will also address next-generation storage mechanisms required by the big data market.”
Exascale computing has applications for many government functions, including espionage. It can help crack encrypted codes, test the vitality of nuclear weapons, model climate change and study disease eradication.
“DOE’s compelling science, energy assurance and national security needs will require a thousandfold increase in usable computing power, delivered as quickly and energy-efficiently as possible,” the solicitation stated. “Funding for the DOE exascale program has not yet been secured, but DOE has compelling real-world challenges that will not be met by existing vendor roadmaps.”
According to experts, energy consumption is perhaps the biggest barrier to reaching exascale capability, particularly memory’s drag on power.
Nvidia chief scientist Bill Dally explained the power problem in a blog post after earning funding: “Theoretically, an exascale system could be built with x86 processors today, but it would require as much as 2 gigawatts of power — the entire output of the Hoover Dam.”
The government seeks to pare down exascale energy consumption to 20 megawatts by the end of the decade, department officials said. Proposals, therefore, were required to offer power-saving designs for, among other things, energy-efficient hardware, novel cooling systems and “power-aware” algorithms.
During the two-year program, government scientists at seven national laboratories will partner with industry researchers across the country.
Energy officials said they recognize that many federal applications of exascale computing involve strenuous computations, data movement and reliability requirements. So the project also aims to support commercial, nonsupercomputing applications that will make the effort financially worthwhile. “Offerors shall describe impact specifically on the [high-performance computing] market as well as the potential for broad adoption,” the contracting documents stated. “Solutions that have the potential for broader adoption beyond HPC are highly desired.”
Large businesses will be required to pay for at least 40 percent of the total value of their work if they want intellectual property rights. For example, the solicitation stated, “If a large business offeror proposes a scope of work with a value of $10 million over two years, and if that offeror wishes to own any intellectual property created in the course of performing the subcontract, the offeror must be willing to accept a subcontract for only $6 million.”
Small businesses, nonprofit organizations and universities can claim the rights to their inventions at no cost.
Exascale computing “will certainly have many derivative benefits to all of us,” Patterson said. “This is a very, very exciting program.”
Lawrence Livermore will run the project under the auspices of the Energy Office of Science and the National Nuclear Security Administration, a department agency that manages the nation’s nuclear weapon stockpile.