It’s that time of the year again: TOP500 published its twice-annual list of the 500 fastestsupercomputers in the world today, and it looks like China’s Tianhe-2 is still on top — and for the sixth consecutive time. There are two new systems in the top 10: At number six is Cray’s Trinity, jointly developed by the Department of Energy’s Los Alamos and Sandia national laboratories, and at number eight is Hazel-Hen, also built by Cray and residing at the HLRS – Höchstleistungsrechenzentrum Stuttgart in Germany.
More telling is what constitutes the rest of the group. The U.S. has the lowest number of systems in the top 500 since the latter was first created in 1993, while China now has almost three times as many systems on the list as it did before. Six of the top 10 systems were installed in 2011 or 2012, the report said, while Tianhe-2 fired up in 2013.
So, onto the results: Tianhe-2 sits on top of the Linpack benchmark with 33.86 petaflops/s (or quadrillions of calculations per second, or Pflop/s). Next up is the Cray XK7 system Titan (pictured below), located at the DOE’s Oak Ridge National Laboratory in the U.S., with 17.59 Pflop/s on the Linpack benchmark.
Trinity, the Cray XC system landing at number six, scored 8.1 Pflop/s with its 301,056 cores, while the 185,088-core Hazel-Han in Germany scored 5.6 Pflop/s. Cray is on a bit of a tear lately, after having previously fallen from favor; it now has 24.9 percent of installed total performance across the TOP500, up from 24 percent, and is the leading vendor in this benchmark overall. IBM is in second place with 14.9 percent, down from 23 (see below as to why), while HP is at third with 12.9 percent (down from 14.2).
Breaking down the list further, the U.S. now contains just 200 of the 500 systems, down from 231 in July and as mentioned the lowest since the list was first created 22 years ago. Next up is now China, leapfrogging all of Europe with 109 systems; the latter now has just 108 in total, down from 141. China can count part of its success to a rebranding of machines from IBM to either IBM/Lenovo or Lenovo/IBM, while the Chinese vendor Sugon has a total of 49 systems in the list.
Another key change: The growth rate is lagging, with a new normal of a 55 percent improvement in performance year-over-year for the past six years, compared with 90 percent year-over-year jumps before that time. Eighty of the 500 systems are faster than 1 Pflop/s. Within the top 10, Tianhe-2 (#1) and Stampede (#10) use Intel Xeon Phi processors, while Titan (#2) and Piz Daint (#7) use Nvidia GPUs to improve computational speeds.
Here’s the current list:
1. Tianhe-2: TH-IVB-FEP Cluster; National Super Computer Center in Guangzhou, China; 3.12 million cores (33.86 petaflop/s).
2. Titan: A Cray XK7 system at the Department of Energy’s Oak Ridge National Laboratory (17.59 petaflop/s).
3. Sequoia: An IBM BlueGene/Q system located at the Department of Energy’s Lawrence Livermore National Lab in California, with 1.57 million cores.
4. K Computer: A SPARC64 system with 705k cores at RIKEN Advanced Institute for Computational Science in Japan.
5. Mira: IBM BlueGene/Q; DOE/SC/Argonne National Laboratory, US; 786k custom IBM cores.
6. Trinity: Cray XC40; DOE/NNSA/LANL/SNL, US; 301,056 Xeon E5-2698v3 cores.
7. Piz Daint: Cray XC30 with 116k Xeon and Nvidia cores; located at the Swiss National Computing Centre in Switzerland.
8. Hazel Hen: Cray XC40; HRLS-Höchstleistungsrechenzentrum Stuttgart, Germany; 185k Xeon E5-2680v3 cores.
9. Shaheen II: A Cray XC40 at King Abdullah’s University of Science and Technology in Saudi Arabia, marking the first appearance of a Middle East supercomputer in the top 10 (5.536 petaflop/s).
10. Stampede: A Dell PowerEdge C8220 system with 462k Xeon Phi cores at the Texas Advanced Computing Center/University of Texas in the US.
The next step is whether anyone can figure out how to build a real quantum computer. IBM has previously said that if someone can build one with just 50 quantum bits (qubits), and that can detect both types of errors and scale to large systems, no combination of today’s TOP500 supercomputers could successfully outperform it.”
Source :ExtremeTech
0 comments:
Post a Comment