Cray's XK6 Hybrid Supercomputer Will Combine AMD, Nvidia Chips

Cray is unveiling the latest of its hybrid supercomputers, which will combine the latest chip technologies from Advanced Micro Devices and Nvidia with Cray’s own Gemini interconnect technology to create a system that offers up to 50 petaflops of compute power.

Cray officials introduced the XK6 supercomputer May 24 at the company’s user group meeting in Fairbanks, Alaska. The system uses AMD’s soon-to-be-announced Opteron 6200 “Interlagos” chips—which offer up to 16 cores per chip and a discrete-level graphics technology integrated onto the same piece of silicon with the CPU—and Nvidia’s Tesla 20-Series GPUs (graphics processing units).

A unified programming environment that includes both AMD’s x86-based chips, Nividia’s GPUs and Cray’s Linux Environment offers organizations with HPC (high-performance computing) environments a tightly integrated supercomputer that can meet their increasing performance demands, according to Barry Bolding, vice president of Cray's product division.

"Cray has a long history of working with accelerators in our vector technologies,” Bolding said in a statement. “We are leveraging this expertise to create a scalable hybrid supercomputer—and the associated first-generation of a unified x86/GPU programming environment—that will allow the system to more productively meet the scientific challenges of today and tomorrow.”

The accelerator-based XK6 system, with its tightly integrated x86-GPU environment, and coupled with the Gemini high-performance networking technology is the latest step forward in Cray’s efforts to meet the needs of HPC environment, he said.

“We built the world's first production petaflops system with the Cray XT5 supercomputer, reinvented high performance networking with the Gemini interconnect, and we are now redefining accelerator-based supercomputing with the unified GPU and scalar technologies built into the Cray XK6 system," Bolding said.

HPC organizations are increasingly turning to systems that include GPU technologies as a way of ramping up performance and parallel-processing capabilities while keeping down power and space costs. AMD is rolling out chips under its Fusion banner, in which the graphics, compute and interconnect capabilities are integrated onto the same die. A growing number of systems vendors, including IBM and Appro, are offering hybrid servers to HPC organizations that run both x86 chips and Nvidia’s Tesla 20-Series GPUs.

For developers, Cray’s new system will include a unified x86/GPU programming environment that will include tools, libraries, compilers and third-party software, according to the company.

The system is expected to be available in the second half of 2011, and organizations will be able to configure it in a single cabinet with tens of compute nodes, or a multiple-cabinet system with tends of thousands of compute nodes, with up to 50 petaflops (or 50 quadrillion floating point operations per second) of performance.

Cray already has a customer for the XK6 system—the Swiss National Supercomputing Center, which is upgrading its Cray XE6m system, nicknamed “Piz Palu,” to a multi-cabinet XK6 supercomputer. The center’s systems support a wide range of work by the Swiss research community in such areas as weather forecasting, chemistry, physics, genetics, experimental medicine, astronomy and computer sciences.