Supercomputer maker Cray will bring together CPUs
from Advanced Micro Devices and graphics technology from Nvidia to
upgrade the massive Jaguar system at a national laboratory and create
what could become the world’s more powerful supercomputer.
Cray officials in May unveiled its latest hybrid CPU-GPU supercomputer, the XK6,
which will be powered by AMD’s upcoming 16-core Opteron 6200
“Interlagos” chips and Nvidia’s Tesla20 Series graphics processing
units (GPUs), and will include Cray’s own Gemini interconnect
technology.
On Oct. 11, Cray executives announced that they
will upgrade the Jaguar supercomputer at the Oak Ridge National
Laboratory in Tennessee, turning Jaguar into what will be known as
Titan, a system with a peak performance of 10 to 20 petaflops—a
petaflop equals 1 quadrillion floating point calculations per second—of
power, according to Cray.
By comparison, the world’s fastest supercomputer, Japan’s K system,
has a peak performance of almost 8.8 petaflops. Jaguar had been the
world’s fastest until last year. Cray officials described the XK6 as a
tightly integrated supercomputer that will includes the CPUs, GPUs,
Gemini interconnect and high-performance software. The result will be a
system that enables researchers at the Oak Ridge lab to throw massive
amounts of computer power at the most complex problems dealing with
energy and environmental issues.
“The new system will enable even further amazing
scientific achievements,” Cray President and CEO Peter Ungaro said in a
statement.
Ungaro added that Titan will be a key step in the
company’s stated goal of scaling the peak performance of the XK6 to
more than 50 petaflops.
High-performance computing organizations
increasingly have been turning to hybrid CPU-GPU supercomputers to
leverage the processing and parallel computing capabilities of graphics
chips. In addition, chip makers AMD and larger rival Intel also are
integrating high-level graphics technologies and CPUs on a single piece
of silicon to boost the performance of their processors while driving
down power consumption.
Cray’s announcement came a day before the Los
Alamos National Laboratory’s Institutional Computing Program announced
it is deploying a supercomputer from Appro—the ApproXtreme-X—for
scientific research around such areas as the oceans, plasma physics and
nuclear energy. Codenamed “Mustang,” the system powered by AMD’s
12-core Opteron 6100 processors, includes 38,400 processing cores and
can reach a performance of 353 teraflops.
Cray’s $97 million project with the Oak Ridge
facilities will be done in phases, with the first phase including
replacing the current XT5 compute blades with Cray’s XK6 blades, which
will comprise the AMD Interlagos chips, the Gemini interconnect and a
subset of XK6 nodes powered by Nvidia’s Tesla GPUs. Cray officials said
they expect to complete the first phase of the project by the end of
the year, despite a delay by AMD in shipping the Interlagos chips.
AMD announced in late September that problems with
manufacturing partner Globalfoundries’ fabrication plant in Dresden,
Germany, had delayed the shipment of Interlagos chips, which happened later in the third quarter than was expected.
The second phase of Cray’s projects will include
equipping the Titan system with Nvidia’s next-generation Tesla “Kepler”
GPUs, and should be finished in the second half of 2012. Upgrades
beyond the two phases could increase the total value of the contract,
according to Cray.
IT & Network Infrastructure News & Reviews 
