RIKEN and Fujitsu have just announced that the "K computer" HPC system that is currently under joint development has achieved a LINPACK benchmark performance of 10.51 petaflops, improving its previous performance of 8 petaflops reached previously this year.
Even before this new result, the Japanese system was already the fastest supercomputer in the world, but now it has reached its final configuration which is comprised of 864 racks including a total of 88,128 interconnected CPUs.
The processors installed are specially tuned
Fujitsu Sparc64 VIIIfx chips with eight processing cores, a TDP of 58W and a peak performance of 128 GFlops at their 2GHz clock rate.
In order to achieve this results Fujitsu has included a set of extensions developed for high performance clusters which allow applications to manage the CPUs 6MB shared L2 cache.
These instructions also provide support for SIMD, 256 floating point registers per core and carry advanced inter-core hardware synchronization capabilities.
In the K Computer, the chips are linked together with a special interconnect called Tofu that is a 6-D mesh/torus capable of providing 5 GBytes/s of bandwidth.
"I am thrilled that the K computer has achieved a
LINPACK benchmark performance of 10 petaflops, while still being in development. I am also proud that we are one step closer to our goal of building a world-class
supercomputer,” said Masami Yamamoto, President of Fujitsu Limited.
“At Fujitsu, we will continue to work tirelessly to deliver the K computer and its system software by June 2012, and enable it to be used by a large number of researchers throughout the world.
“We truly hope that the K computer's world-class performance will benefit and deliver value to both Japan and the entire world with leading-edge computing technology," concluded the company's rep.
The scientists expect the K Computer to deliver even more performance in the future after they adjust the system's software.
Find us on Google+
No user comments yet.
Be the first to express your opinion!
