The U.S. Department of Energy's (DOE) Oak Ridge National Laboratory on Monday launched Titan, a system capable of churning through more than 20 000 trillion calculations each second – or 20PFLOPS - by employing both high-performance multi-core AMD Opteron microprocessors and Nvidia Tesla “Kepler” compute accelerators. Titan is ten times more powerful than Jaguar, a previous-gen supercomputer.
The Cray XK7 system contains 18688 nodes, with each holding a sixteen-core AMD Opteron 6274 processor and an Nvidia Tesla K20 graphics processing unit (GPU) accelerator (based on GK110 processors). Titan has more than 710TB terabytes of memory and uses Cray’s high-performance Gemini interconnections. The Cray XK7 system is capable of scaling to more than 50PFLOPS of performance. Titan was upgraded from a Cray XT5 supercomputer named “Jaguar”. The combination of central processing units, the traditional foundation of high-performance computers, and more recent GPUs will allow Titan to occupy the same space as its Jaguar predecessor while using only marginally more electricity, yet delivering ten times higher performance.
By relying on its 299 008 x86 microprocessor cores to guide simulations and allowing its new Nvidia GPUs to do the heavy lifting, Titan will enable researchers to run scientific calculations with greater speed and accuracy. Titan will provide unprecedented computing power for research in energy, climate change, efficient engines, materials and other disciplines and pave the way for a wide range of achievements in science and technology.
The Cray XK7 supercomputer also features a unified CPU/GPU programming environment that provides users with validated tools, libraries, compilers and third-party software, fully integrated with the system's hardware. When combined with the Cray Linux Environment, the result is a hybrid supercomputer that blends scalable hardware, software and network. Cray XK7 customers will be able to utilize the capabilities of a multi-purpose supercomputer designed for the next-generation of many-core, HPC applications.
"One challenge in supercomputers today is power consumption. 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," said Jeff Nichols, associate laboratory director for computing and computational sciences.
Jaguar Becomes Titan
The transformation from Jaguar to Titan is another significant milestone in the collaborative partnership between Cray and ORNL that has produced groundbreaking HPC accomplishments. In 2008, Jaguar set a world record for computer speed with sustained performance of more than one petaflops on two scientific applications, and the system subsequently passed that threshold a total of five times on real-world applications. In 2009, Jaguar claimed the number one spot on the list of the fastest supercomputers in the world. In October 2011, Cray announced it had received a contract to upgrade Jaguar to Titan and equip the system with Nvidia Tesla compute cards. Today, the Cray XK7 system made its debut.
“Today's unveiling of the Titan supercomputer is an exciting moment for Oak Ridge and the Department of Energy's Office of Science, and while the system is currently going through the acceptance process, all of us at Cray share in the enthusiasm that surrounds this amazing tool for open science. The Titan supercomputer is an incredibly powerful Cray XK7 system combining innovative technologies from companies such as AMD and Nvidia, surrounded by a tightly-integrated Cray hardware and software infrastructure. With today's launch of the Cray XK7, we can now offer our customers the same technologies found in one of the most powerful supercomputers in the world,” said Peter Ungaro, president and chief executive of Cray.
Upgradeable from Cray XT4, Cray XT5, Cray XT6, Cray XE6 or Cray XK6 systems, the Cray XK7 supercomputer is available now. The system can be configured in a single cabinet with tens of compute nodes, to a multi-cabinet system with tens of thousands of compute nodes.