IBM announced BlueGene/Q, its next-generation supercomputer
architecture, which will debut in the Sequoia system next year and
eventually scale to more than 100 PFlops – more than ten times the
performance of today’s fastest supercomputer.
IBM BlueGene/Q
The final version of Sequoia, deployed at Lawrence Livermore National
Laboratory (LLNL) is expected to reach about 20 PFlops in next year and
become one of the world’s fastest supercomputers as well as the world’s
most efficient supercomputer with a performance of 2 GFlops per watt.
Sequoia system will integrate 1,572,864 processing cores (98,304 16-core
PowerPC A2 processors) in 96 racks.
However, BlueGene/Q is especially interesting as IBM today confirmed
that Bluegene/Q is capable of more than 100 PFlops, which was described about two months ago in a patent application submitted
to the USPTO. According to that filing, a massive document with 649
pages and 2263 claims, a 100 PFlops BlueGene/Q system could consist of
1024 compute node ASICS in 512 racks – representing a total of 524,288
nodes and 8,388,608 processing cores.
According to IBM, each processor consumes 30 watts of power, which
puts Sequoia power consumption including storage and cooling
requirements into the neighborhood of about 6 to 8 MW, while a 100
PFlops system will easily exceed 30 MW. In comparison, today’s fastest
supercomputer, Japan’s K Computer, delivers 10 PFlops via 705,024
Sparc64 VIIIfx processors for about 12.7 MW.
Wolfgang Gruener in Business Products on November 15
No comments:
Post a Comment