The Department of Energy's Oak Ridge National Laboratory (ORNL) announced that they have awarded a contract to Cray Inc. to increase the Jaguar supercomputer's science impact and energy efficiency. The upgrade, which will provide advanced capabilities in modeling and simulation, will transform the DOE Office of Science-supported Cray XT5 system, currently capable of 2.3 million billion calculations per second (petaflops), into a Cray XK6 system with a peak speed between 10 and 20 petaflops.
ORNL's Jaguar took the Top500 number one spot in November 2009 and was bumped off the top a year later. As of June 2001 it was ranked third. Jaguar is a mix of Cray XT5 and XT6 cabinets with 224,256 cores and 362TB of main memory, and as The Register puts it, will change its spots from CPUs to GPUs with the upgrade. The resulting supercomputer, named Titan, will use the latest AMD Opteron CPUs as well as NVIDIA Tesla graphics processing units. Titan will have 299,008 cores and 600 terabytes of memory and is expected to be completed in late 2012 and ready for users in early 2013.
"Titan will allow for significantly greater realism in models and simulations and the resulting scientific breakthroughs and technological innovations will provide the return on this national investment," said ORNL Director Thom Mason. "Discoveries that take weeks even on a system as powerful as Jaguar might take days on Titan."
The first part of the upgrade will take the system from two six-core processors per node to one 16-core processor per node, and facilitate the addition of GPUs. "These Tesla GPUs are able to perform many more calculations for the same amount of power as conventional microprocessors and will work hand-in-hand with CPUs to deliver new levels of energy-efficient application acceleration," said Mason.
The first phase of the upgrade project will begin later this year and include 1,000 current NVIDIA Tesla 20-series GPUs. In the second phase, to be performed in the second half of 2012, between 7,000 and 18,000 of NVIDIA's next-generation Tesla GPUs will be added to the system, increasing the peak performance to between 10 and 20 petaflops, depending on architectural options executed.
The Cray XK6 supercomputer was introduced earlier this year and set out to combine their Gemini interconnect, AMD Opteron 6200 Series processors (code-named "Interlagos") and NVIDIA Tesla 20-Series GPUs to create a tightly-integrated supercomputing system upgradeable to more than 50 petaflops (quadrillions of operations/second) of compute power.
"Cray has been fortunate to share in a great partnership with the Oak Ridge National Laboratory and the Department of Energy's Office of Science, and we are excited that our relationship will continue with Titan," said Peter Ungaro, president and CEO of Cray. "Scientists at Oak Ridge and across the DOE have been researching some of the world's most pressing energy and environmental challenges. As a company, it's an honor to know that this vitally important work has been, and will now continue to be, done on supercomputers built by Cray. It's a partnership we are very proud of."
Exascale and Beyond
NVIDIA's Steve Scott discussed the Titan Supercomputer announcement recently and how it will be an important milestone on the way to Exascale computing. Exaflop systems will be 1,000 times faster than petaflop systems, delivering one billion billion calculations per second. Steve says that while plenty of computing challenges have been outlined for exascale-class computing to solve, the exascale computer using today's x86 technology would require two gigawatts of power. This is where GPUs have the potential to help solve the power problem, as GPUs are designed to run many parallel tasks as power-efficiently as possible. NVIDIA co-founder, president and CEO Jen-Hsun Huang will deliver the keynote address at the SC11 event in Seattle next month.