Wednesday, November 23, 2016

IBM to deliver two hundred-petaflop supercomputer with the aid of early 2018

greater supercomputer news this week: the us is responding to China’s new Sunway TaihuLight system that become announced Monday, and speedy. First, the department of power’s (DOE) all rightRidge national Laboratory is anticipated to take transport of a brand new IBM machine, named Summit, in early 2018 so one can now be able to 2 hundred peak petaflops, Computerworld reviews. that would make it nearly twice as fast as TaihuLight if the declare proves authentic. (We had at the beginning stated in 2014 that both Summit and Sierra would obtain more or less 150 petaflops.)
TaihuLight (pictured below) now sits at primary on the two times-every year TOP500 list of the quickest supercomputers within the international, with a Linpack benchmark score of 93 petaflops and a claimed height of 124.five petaflops. The ultra-modern TOP500 statement Monday prompted a piece of a stir. no longer most effective is TaihuLight more or less three times faster than China’s Tianhe-2, the earlier champion, but it also makes use of no US-sourced parts in any respect for the first time, because it’s powered by using Sunway 260-core SW26010 processors which might be more or less on par with Intel Xeon Phi, as well as custom proprietary interconnect.

In turn, Summit will appoint IBM Power9 and Nvidia Volta GPUs. Summit will deliver over five instances the computational overall performance of Titan’s 18,688 nodes using most effective approximately three,four hundred nodes. every node could have “over half a terabyte” of so-known as coherent memory (HBM + DDR4), plus 800GB of non-unstable RAM that serves as a burst buffer or extended reminiscence.

Titan (pictured below), meanwhile, is currently #three at the TOP500 list, and resides at the DOE very wellRidge countrywide Laboratory, in which the new IBM Summit system could be positioned. In a announcement through Computerworld, the DOE pointed out that considering 1993, US supercomputing competencies “have grown exponentially by way of a factor of 300,000,” and that “high-overall performance computing stays an essential precedence for the DOE.” (For more on alrightRidge’s earth, strength, weather, and geographic science research the usage of those systems, go to its devoted supercomputing page.

The Linpack benchmark has emerged as a novel yardstick for measuring the performance of complex supercomputers. It doesn’t record ordinary overall performance in all situations; as an alternative, it measures the overall performance of a system while solving a “dense gadget of linear equations” that gives a terrific approximation of real-international (rather than top) overall performance. examine extra approximately the benchmark at the above link.

one after the other, Cray introduced this week at the 2016 worldwide Supercomputing convention in Frankfurt, Germany that its Cray XC structures are now to be had with the modern Intel Xeon Phi (Knights landing) processors. The employer stated the brand new XC systems, which function an adaptive design that supports a couple of processor and storage technologies in the equal architecture, deliver a a hundred% performance increase over previous generations. Cray also unveiled the Sonexion 3000 Lustre garage device, that can supply speeds of just about 100GB/sec in a unmarried rack.

No comments:

Post a Comment