Titan: The World’s Fastest Supercomputer Unleashed
Michael Harper for redOrbit.com – Your Universe Online
The Oak Ridge National Laboratory (ORNL) has created a machine capable of doing the unthinkable. In fact, it’s capabilities not only sound impossible, they sound fictional. By combining traditional CPUs with graphical processing units (GPUs), the Titan Supercomputer capable of spitting out 20 petaflops, making it 10 times more powerful than ORNL’s previous Jaguar supercomputer.
For the uninitiated, a petaflop, says Buddy Bland, project director at ORNL, is a way to express trillions of calculations per second. 20 petaflops, as it turns out, means 20,000 trillion calculations every second. By partnering CPU and GPU, ORNL has built an even more powerful machine, capable of even more petaflops and also creating 3D simulations of cells, molecules and the way drugs interact with them.
Running some 200 refrigerator sized-units in an area the size of a basketball court can rack up a hefty electrical bill, and as such the Titan supercomputer is supported by the Department of Energy (DoE) who will take advantage of all these petaflops to conduct research in climate change, efficient engines, energy and the like. The DoE has been operating supercomputers at the ORNL since 2005, continually upgrading as technological advancements come along, Titan isn’t so much a new computer as it is an upgrade of the existing Jaguar supercomputer, which first came online in 2005. Up through June 2010, the Top 500 continued to list Jaguar as the world’s fastest supercomputer, only to be beaten out by China’s Tianhe-1A just six months later, knocking Jaguar to number 2.
With these upgrades, Titan will once again place ORNL and the DoE at the top of the supercomputer heap.
The infrastructure of Titan is built upon the Cray XK7 system, which houses 18,688 nodes, according to Jeff Nichols, associate laboratory director for computing and computational sciences at ORNL, in a press release.
Each of these nodes contains both a 16-core AMD Opteron 6276 processor and an NVIDIA Tesla K20 GPU accelerator, enough to make even the most powerful desktop look like a solar powered calculator. Titan also has an impressive brain, measuring 700 terabytes large. Though operating Titan is no easy task (a combination of fans and water pipes are used just to keep the system cooled), Nichols says the combination of CPU and GPU makes running Titan even more efficient than before.
“One challenge in supercomputers today is power consumption,” said Nichols in the press statement.
“Combining GPUs and CPUs in a single system requires less power than CPUs alone and is a responsible move toward lowering our carbon footprint. Titan will provide unprecedented computing power for research in energy, climate change, materials and other disciplines to enable scientific leadership,” explained Nichols.
This marriage of CPU and GPU will also enable the scientists and researchers at ORNL to handle hundreds of calculations simultaneously, delegating the simulations to the CPUs while the GPUs take care of the real heavy work. The end goal of such a computational behemoth, of course, is to run simulations as a part of larger experiments.
“Titan will allow scientists to simulate physical systems more realistically and in far greater detail,” said James hack, the director of ORNL’s National Center for Computational Sciences.
“The improvements in simulation fidelity will accelerate progress in a wide range of research areas such as alternative energy and energy efficiency, the identification and development of novel and useful materials and the opportunity for more advanced climate projections,” added Hack.
Titan will be available to other teams for their projects, final approval is left up to Cray and ORNL. The DoE has the first shot at Titan, of course, and will use the computer as a part of their Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program.