Not logged in » Login

Recent Discussion:

NunoCosta | 02.06.2020, 17:25
Hi mlegrafleitas.Thank you for your question.We are working on providing you the best answer, I'll g...
mlegrafleitas | 01.06.2020, 17:28
Hi everyone,Need your help to compare this 2 Storage from FUJITSU and DELL because I have a customer...
HarryD | 24.03.2020, 08:46
Thank you all for helping out.The problem has been solved.
Jun 22 2017

Swiss System Shakes Up TOP500 List of Supercomputers

HPC enthusiasts on both sides of the Atlantic may not enjoy hearing this, but their renowned biannual ranking of the world's most powerful scientific computer systems sometimes suffers from the same key symptom that makes many soap operas so hard to digest: predictability. Once a supercomputer has conquered a spot among the 10 or 20 fastest number crunchers, it usually retains that position for years – much like Larry Hagman and Patrick Duffy did in "Dallas." The TOP500's latest edition (no. 49) seems to adhere to the same rule, but eventually provides a little more entertainment value.

As could be expected, the very same systems that topped the previous list return to their key roles for the June 2017 ranking published on Monday this week – except for one that now claims an even bigger part. Positions number 1 and 2 are still occupied by China's Sunway TaihuLight and Tianhe-2, which continue to deliver 93 and 33.9 petaflops (quadrillions of floating point operations per second) respectively and are located at the National Supercomputing Centers in Wuxi and Guangzhou. Together, both systems deliver 17.7 petaflops more than the next eight systems on the list – or more bluntly put, than the rest of the top 10 combined, even though they feature the very same setups as they did in June last year, when they first reached their current positions.

The system providing the entertainment factor is the new 'bronze medalist' Piz Daint, which resides at Switzerland's Centro Svizzero di Calcolo Scientifico (CSCS) in Lugano. Originally a Cray XC30 system consisting of 5,272 compute nodes built from 8-core Intel® Xeon® E5-2670 processors and NVIDIA® Tesla® K20X graphics cards, it first hit the TOP500 list at position number 6 in November 2013 with a Linpack performance of 6.27 petaflops. Meanwhile, it has been upgraded several times and 'fused' with its twin system Piz Dora, resulting in a hybrid XC50/XC40 system holding over 30,000 third-generation, 12-core Intel® Xeon® E5-2690 processors and a slew of NVIDIA Tesla P100 GPUs that serve as compute accelerators. Last November, this setup scored a respectable 9.8 petaflops and came in eighth; the latest rearrangement, however, brought about another significant boost in GPU capacity that lifted the overall performance to 19.6 petaflops – enough to push aside Titan, a Cray XK7 system run by the U.S. Department of Energy at its Oak Ridge National Laboratory. This 5-position-jump among the top 10 may already seem unusual enough, but even more important is the fact that this rare occurrence marks only "the second time in the 24-year history of the TOP500 list that the United States has failed to secure any of the top three positions," according to the official press release – the other occasion being in late 1996, when three Japanese systems outran all their competitors.

Speaking of Japanese successes, it should be noted here that Fujitsu still holds its own among the leading forces in supercomputing – with two completely different installations. The near-legendary K computer, which is based on the company's own SPARC processors and has been a regular among the fastest systems since June 2011, came in at number 8 with a performance of 10.5 petaflops – an excellent score for a six-year-old system. However, it was outpaced by Oakforest-PACS, a configuration built from more than 8,200 off-the-shelf PRIMERGY CX1640 M1 server nodes, each of which is equipped with a 68-core Intel® Xeon Phi™ 7250 processor. The entire setup comprises nearly 560,000 cores along with 919 TB of main memory and Intel's new Omni-Path Interconnect and scores a Linpack performance of 13.6 petaflops, which means it ranks 7th on the TOP500 list. In addition, its theoretical peak performance could hit 24.9 petaflops – which is nearly on par with that of Piz Daint. Oakforest-PACS is operated by the Joint Center for Advanced High Performance Computing, a collective effort of the Tokyo and Tsukuba universities. Its basic specification (PDF) is available here.

 
SHARE

Comments on this article

No comments yet.

Please Login to leave a comment.

X

Please login

Please log in with your Fujitsu Partner Account.

Login


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now