Michael Disabato, a vice president and service director at the Burton Group, says this kind of advancement will be critical to a select number of companies. "If you're a financial services company and want to predict the markets, it's going to mean a lot," he adds. "If you're at NASA trying to model the increasing number of failure modes for the space shuttle, you probably want one of these. But if you're in a hospital working on patient care, you probably won't need one of these."
But CIOs eager to get their hands on one of these supercomputers better pull out their checkbooks and make a whole lot of room.
Brandt says the initial base cluster for Roadrunner will cost about $35 million. He doesn't know at this point how much it will cost once the Cell chips are added. And if companies can afford that kind of price tag, they have to find room for the approximately 360 racks of servers they'll have to house, power, and cool. IBM expects it will take up 12,000 square feet of room--or about the size of three basketball courts.
The teraflop speed barrier was broken in 1997. A teraflop is 1 trillion calculations per second. A petaflop is a thousand times that speed, or a quadrillion, which is a one followed with 15 zeros. At this point, Blue Gene is still the fastest supercomputer on the block. Installed at the Lawrence Livermore National Laboratory in Livermore, Calif., the machine is based on IBM's Power chip. It peaks at more than 360 teraflops. But when Roadrunner gets the Cell chip addition, it should be four to eight times faster than Blue Gene is today.
Don't count Blue Gene out, though, says Brandt. They're still working to speed it up and expect it to pass the petaflop barrier as well.
Brandt says he's looking forward to having both supercomputers pass the petaflop barrier.
"They're both capable of hitting a petaflop and beyond," says Brandt, adding that Blue Gene is ideal for life sciences calculations, while Roadrunner will be better suited to run calculations for weapons codes, along with seismic and weather forecasting. "It's not a race."