Cray plans massive supercomputer upgrade

Posted by TG Daily Staff

Cray has clinched a lucrative deal to upgrade the XT5 supercomputer - aka "Jaguar" - located at the Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL).

Dubbed "Titan," the new XK6 is expected to boast a peak performance between 10 and 20 petaflops (quadrillion mathematical calculations per second) of high performance computing (HPC) power.

Cray plans massive supercomputer upgrade"The Titan system at ORNL will bring together the features of a proven, production petascale architecture with innovative Nvidia Tesla graphic processing unit (GPU) technologies to create a supercomputer capable of unprecedented scale," explained Cray CEO Peter Ungaro.

"[The] XK6 will feature productive, high performance software that leverages a proven, scalable system interconnect and a powerful blend of GPUs and general purpose central processors (CPUs) in a single, tightly integrated supercomputer."



According to Ungaro, the multi-year, multi-phase contract is valued at more than $97 million. 

The first phase of the contract will include replacing the Cray XT5 compute blades with Cray XK6 compute blades, which features AMD's Opteron processors code-named "Interlagos," Cray's Gemini interconnect, and a subset of Cray XK6 nodes equipped with Nvidia Tesla 20-series GPUs. 



The second phase of the contract - equipping the system with Nvidia Tesla GPUs based on the next-generation architecture code-named "Kepler" - is slated to be completed in the second half of 2012. 



Launched in May 2011, the Cray XK6 system is a hybrid supercomputer that integrates Cray's Gemini interconnect, AMD Interlagos chips and Nvidia Tesla 20-Series GPUs. 



The unified CPU/GPU programming environment offers a range of tested tools, libraries, compilers and third-party software that is fully integrated with the system's hardware. By combining AMD processors with Nvidia GPUs, Cray XK6 customers can run apps with either scalar or accelerator components, resulting in one of the first general-purpose GPU supercomputers capable of more than 50 petaflops of performance.