They’re building the future at IBM in Rochester, Minn. – and her name is Mira.
Mira is an IBM supercomputer worth an estimated $50 million, and when she makes her debut next year, IBM says she may be the most powerful computer in the world.
Mira will produce more than 10 quadrillion calculations per second, IBM promises. That’s four times faster than China’s Tianhe-1A, currently considered the fastest.
If everyone in the United States made one calculation per second, it would take them nearly a year to complete what Mira can do in one second.
We’re not out just to prove that our supercomputer is No. 1, said Mike Good, software developer for advanced systems at the Rochester plant. We want to come out with systems that are innovative but also have the ability to solve the world’s problems in a real way.
IBM supercomputers are used to design new drugs, develop nuclear weapons and help search for oil. Once completed, Mira will handle tasks such as designing advanced electric car batteries.
IBM announced last week that it has received a contract from the U.S. Department of Energy to build Mira at its Rochester facility – one of the last vestiges of Minnesota’s heady supercomputer past, when Cray Research and Control Data Corp. were based in the Minneapolis area.
IBM declined to reveal Mira’s purchase price, which comes out of a $180 million budget at Argonne National Laboratory in Chicago, where Mira will end up after it’s built. Rich Brueckner of Portland, Ore., who runs the InsideHPC.com (high performance computing) news website, estimates Mira will cost about $50 million, plus another $4 million to $5 million annually to pay for her electricity bills.
Mira is what IBM calls a Blue Gene/Q supercomputer, the latest in a series IBM has been building since 1999. She is made up of a lot of little parts, including 786,000 individual processing units.
Given the frantic pace at which technology advances, Mira may or may not be the fastest in the world when the Top 500 list of supercomputers is determined by university and government laboratory experts at the end of 2012. It just depends on what supercomputers other nations and companies build between now and then, analysts say.
IBM says it doesn’t matter. The important thing, company officials emphasize, is scientific research, although they acknowledge value in being able to tout the world’s fastest computer.
Analysts say measurements used to pick the fastest computer are of dubious value. They say Mira’s real contributions will be everything IBM learns while building her. Such innovations will trickle down to all the lower-priced computers that IBM sells to corporations, most of which don’t need a $50 million supercomputer, but need something fast.
Wall Street will use those computers to analyze their portfolios to see what will happen if interest rates do this or that, said Dan Olds, an analyst at Gabriel Consulting Group in Beaverton, Ore.
IBM also sees Mira as a steppingstone to its next generation of supercomputers that will be 1,000 times faster, Good said.
Despite Mira’s great speed, she may never run flat out at 10 quadrillion calculations, or petaflops, a second, because there are only a half-dozen scientific problems in the world that could use that much speed, Brueckner said. More likely Mira will spend her life juggling several weighty scientific problems at a time, none of them able to single-handedly put her speed to the test.
These supercomputers are not anything you or I would ever use, Brueckner said. But there are plenty of high-energy physics problems that need something that big.