Titan, the supercomputer introduced by the U.S. Department of Energy’s Oak Ridge National Lab (ORNL), is on track to be the world’s fastest. At least until the next one comes along. It can handle over 20,000 trillion calculations a second. That’s the equivalent of seven billion people carrying out three million calculations per second. You know what it has to thank for all that power? Computer gaming.
The graphics processing units (GPUs), usually reserved for consumer and entertainment purposes are the same ones that fuel a machine that could potentially save the world. No exaggeration: Titan is capable of everything from predicting climate change and weather events (including monstrous hurricanes) to finding cures for Parkinson’s and cancer. All because of the same little processor that makes Call of Duty look so realistic.
Faster, Better, Stronger
Compared to its predecessor, Jaguar, Titan is faster and 10-times more powerful, while taking up no more physical space. (It does consume about two megawatts more energy). Inside the Titan reside thousands of NVIDIA’s latest GPU accelerator, the Tesla K20. Each of the computer’s 18,688 nodes holds one of these GPUs as well as a 16-core AMD Opteron central processing unit (CPU).
Researchers at NVIDIA found that the processors they created were starting to resemble the ones used for simulating physics. The rapid-fire calculation solving done by GPUs resembled the way supercomputers simulated problems using CPUs. By combining the two, the CPU is able to handle creating the problem, or simulation, while the GPU is what does the heavy calculations. This makes complex calculations happen faster and allows them to cover more scientific ground. Added bonus? Combining the industry standard CPU with a high performance graphics processor not only makes the supercomputer faster, it also makes it more energy-efficient.
Wait, what? It uses more power, how is that more energy efficient?
Power consumption is a big deal for supercomputers these days, and according to Jeff Nichols, associate laboratory director for computing and computational sciences at ORNL, GPU’s could hold the key to improving that. “Combining GPUs and CPUs in a single system requires less power than CPUs alone and is a responsible move toward lowering our carbon footprint.” While the Titan consumes enough energy to power a small town, one GPU uses eight times less energy per calculation than a CPU. Titan’s size and power makes the uptick in energy usage well worth it, it does more with less energy than a supercomputer of the same size running on the standard method of CPU-only use would.
Skeptics Be Damned
Steve Scott, CTO of NVIDIA’s Tesla business unit told the Washington Post that this isn’t the first time CPUs and GPUs have been combined, and that a lot of people didn’t think it would make much of a difference. The size and scale of the Titan project made the stakes that much higher: the new supercomputer had to be able to handle six applications in tough areas of research including astrophysics, biofuels and nuclear energy. Scott wanted to prove using GPUs in supercomputers wasn’t a stunt. If people see it that way now, at least it’s one that worked.
The goal wasn’t just to create the biggest and baddest supercomputer, it was also to create a new standard. Building a supercomputer using components that many people use to play games shows just how far consumer technology has come – remaking not just business tech but also the even-more-complex world of research technology.
Image Credit: Oak Ridge National Laboratory .