How to Make a 50 Petaflop Supercomputer Even More Powerful

Supercomputing has hit a wall—researchers can't simply daisy-chain any more processors together to increase computational power due to energy constraints. The new Titan project hopes to do an end-around on that issue by transferring some of the load to...graphics cards?

Yes, Nvidia graphics cards like those in high-end gaming rigs. Their main advantage is that graphics processors are capable of handling more simultaneous tasks while expending less energy than microprocessors. By lowering the amount of energy it takes to do the computations, more processing power can be added using the savings in energy.

Developed and implemented by Oak Ridge National Lab, the Titan project will first upgrade the ORNL's Cray XK6 and then, beginning next year, install as many as 18,000 Nvidia GPUs around its AMD x86 chips. This could make it as much as twice as fast and three times as energy efficient as Japan's K computer, currently the world's fastest.

"To keep getting performance gains in the future," Steve Scott, Nvidia CTO, told the NY Times, "we are going to have to get a lot more energy efficient."

Once the system gets up and running, it's expected to perform research into commercially-viable renewable energies. [NY Times]