The Tesla K10 uses two of the GK104 Kepler GPUs to increase processing speeds. (Nvidia)
Nvidia says its new graphics chip can help turn the rising flood of ISR data into a productive river of information.
Company officials say their Tesla K10 graphics processing unit (GPU) relies on a brand-new architecture with 192 cores, dubbed Kepler, which will allow it to perform twice or thrice as fast as its old 32-core, Fermi-based GPUs.
Nvidia says that the new GPU, to be unveiled at the GPU Technology Conference on May 16 and released for sale later this month, will provide faster signal and intelligence processing, video analytics and stabilized videos that provide object detection, and computations that turn flat satellite images into a representation of the curved earth.
Analysts can also use GPUs to process data from collection platforms such as Humvee-mounted cameras or detectors, UAVs, or street sensors that can then be used for simulations or route planning.
“I think that will translate right down to training and mission planning,” said Kevin Berce, business development manager at Nvidia. “They’re going to have real-time data from the theater to do their simulations on. So the faster that they can get that data so they can have real live training exercises – it’s going to be invaluable.”
CPUs, or central processing units, have long been used to analyze this kind of data, but GPUs are able to offload some of the computations to speed up the process considerably.
“It’s like adding a turbocharger to a car,” said Sumit Gupta, senior director of high-performance GPUs at Nvidia. CPUs will get you there, but GPUs will get you there faster.
Using GPUs to increase computational power can also reduce the size of the system required, allowing for greater portability and local turnaround. Gupta envisioned small IED radars in Humvees that could provide real-time analysis of the surroundings in a matter of milliseconds, where normally soldiers would have to send that data to a center for processing.
“You want real-time, in-the-field analysis,” Gupta said. “The speed of simulation or analytics is very important. If you could do it, but it takes six hours — by which time the object is gone — there’s no value anymore.”
One defense contractor that Nvidia did not name has moved some of its processing algorithms from CPUs to GPUs and saved 25 percent on their costs.
Furthermore, Berce added, that company expects to save 22 percent on power costs — something that, if translated to the field, could save money and lives of troops through reduced fuel consumption.