China’s Milkyway 2, also known as Tianhe-2, has retained the number one ranking on the biannual Top500 list of the most powerful supercomputers in the world for a third consecutive time.
Recording the same 33.86 petaflop/s (quadrillions of calculations per second) on the Linpack benchmark as the last November 2013 list, Milkyway 2 reflects slowing growth of the power the world’s fastest supercomputers.
The latest Top500 was announced as the International Supercomputing conference that gets started in Leipzig, Germany, this week.
The 43rd edition of the Top500 list has only one change in the top 10 systems, with a Cray XC30, installed at an undisclosed U.S. government site, at number 10.
Fastest computer 30 petaflop/s faster than number 10
With a 30 petaflop/s difference between number one and number 10 on the June list, the large installations at the top have stagnated. Combined performance of all 500 systems has grown to 274 Pflop/s, compared to 250 Pflop/s six months ago and 223 Pflop/s one year ago.
A lack of new large-scale installations may be to blame for the lack of growth in Top500 performance ratings, however many other factors contribute.
The race to exascale seemed to derail briefly last year, as the industry talked about HPCG (High Performance Conjugate Gradients) as an effort to create a more relevant metric for ranking HPC systems. The HPCG metric addresses the fact that HPC system designs are no longer driven by pure computational performance alone.
The Green500 list of most energy efficient supercomputers, and Graph500 list of data-intensive supercomputers indicate the importance of supercomputer power consumption and workloads.
Other highlights from the June 2014 Top500 list include:
- U.S. installations, while still the top country overall, are down from 265 to 233. The number of Chinese systems on the list rose from 63 to 76.
- A total of 62 systems on the list are using accelerator/co-processor technology, up from 53 from November 2013. Milkyway-2 and Stampede (a University of Texas at Austin supercomputer) are using Intel Xeon Phi co-processors, while Titan (Oak Ridge National Lab) and Piz Daint (Swiss National Supercomputing Center) are using NVIDIA GPUs as co-processors.
- Intel processors power 85.4 percent of the Top500 systems.
- The number of systems by vendor ranks HP first, IBM second and Cray third.
New HPC systems and approaches that are being developed may rescue the stagnated Top500 list in the future. Oak Ridge, Argonne and Lawrence Livermore labs are working together on CORAL – a next-generation supercomputer that may reach 100-200 petaflops. Earlier this month HP rolled out Apollo — a new converged HPC system featuring up to 160 servers per rack and 100 percent liquid cooling.