10 of the world's fastest supercomputers

The cream of the Top500 supercomputer list

10 of the worlds fastest supercomputers
Peter Sayer/IDG

10 of the world's fastest supercomputers

The cream of the Top500 supercomputer list in November 2017, these 10 supercomputers are used for modeling the weather, weapons, ocean currents and other physical phenomena. They not only outperform every other machine on the planet, they also demonstrate the technologies your business might use, at a smaller scale, to get the best performance in the least space or using the least energy.

10. K Computer

K Computer, built by Fujitsu for Japan's RIKEN Advanced Institute for Computational Science (AICS), was the world's fastest computer in June 2011, but by November 2017 had slipped to 10th place in the Top500 list of of the world's fastest supercomputers.

It's the highest-ranked Sparc-based computer in the list, using 88,128 SPARC64 VIIIfx octa-core processors running at 2GHz, for a total of 705,024 cores. It has a maximum sustained performance of 10.51 petaflops, and a peak performance of 11.28 petaflops.

One thing it has nothing to be proud of: Its performance per watt. At just 0.83 gigaflops per watt, it's the least efficient machine in the top 10.

9. Oakforest-PACS

Another Fujitsu machine, Oakforest-PACS was built for Japan's Joint Center for Advanced High Performance Computing. Based on Fujitsu's Primergy CX1640 cluster and using Intel Omni-Path interconnects, Oakforest-PACS entered the Top500 list in 6th place in November 2016, but had slipped to 9th place a year later. 

Its 8,178 Intel Xeon Phi 7250 processors have a total of 556,104 cores. They deliver a maximum sustained performance of 13.55 petaflops, and a peak performance of 24.91 petaflops.

Its power efficiency is a respectable 4.986 gigaflops per watt. 

8. Cori

Cori is a Cray, a name once synonymous with supercomputing. It belongs to the U.S. Department of Energy and sits in the Lawrence Berkeley National Laboratory. Cori entered the Top500 list in November 2016, one place above Japan's Oakforest-PACS, and has slid down in step with it since. In November 2017, it was in 8th place. 

Another thing it shares with Oakforest-PACS is Intel's Xeon Phi processor. It has 9152 of them, for a total of 622,336 cores, with a maximum sustained performance of 14.01 petaflops and a peak performance of 27.88 petaflops.

Strangely for a computer owned by the Department of Energy, Cori is less power efficient that its Japanese rival, delivering a less impressive 3.558 gigaflops per watt.

7. Trinity

Trinity, built by Cray for the U.S. Department of Energy's Los Alamos National Laboratory, received an upgrade in the Summer of 2017, pushing it a few places back up the Top500 list. 

It entered the Top500 list in 6th place in November 2015, equipped with 18,816 Intel Xeon E5 16-core processors, but steadily lost ground, falling to 10th place over the next two years.

In November 2017, with the addition of 9,984 Intel Xeon Phi 68-core processors, it climbed back to 7th place. The upgrade gives it a total of 979,968 cores, and maximum sustained performance and peak performance of 14.14 petaflops and 43.90 petaflops, respectively. Overall, it consumes a little less power than its stablemate, Cori, and is marginally more efficient, consuming 3.678 gigaflops/watt. 

6. Sequoia

Sequoia was the world's most powerful supercomputer back in June 2012, but has been sliding down the Top500 list ever since. By November 2017 it had slipped to 6th place. 

Like Cori and Trinity, Sequoia belongs to the U.S. Department of Energy, sitting in the Lawrence Livermore National Laboratory.

Unlike them, though, Sequoia was built by IBM and uses that company's 16-core Power BQC chips with a custom interconnect. It has 98,304 processors for a total of 1,572,864 cores, each with 1 gigabyte of RAM. 

Together, they deliver a maximum sustained performance of 17.17 petaflops and a peak performance of 20.13 petaflops. When it comes to energy efficiency, the Power chips Sequoia uses are nothing spectacular, delivering a middling 2.177 gigaflops/watt.

5. Titan

Titan was the computer that toppled Sequoia from the top of the charts when it entered the Top500 list in November 2012, but had slid to 5th place by November 2017.

It's the most powerful of the U.S. Department of Energy's computers, and is sited at Oak Ridge National Laboratory.

Like Cori and Trinity, it is made by Cray -- but unlike them, it uses Advanced Micro Devices' 16-core Opteron 6274 chips with Nvidia Tesla accelerators. It has a total of 560,640 cores, delivering maximum sustained performance of 17.59 petaflops and peak performance of 27.11 petaflops.

When it comes to power efficiency, the 2012 Opteron-Tesla combination is on a par with IBM's 2012 Power BQCs, delivering 2.143 gigaflops per watt.

gyoukou supercomputer

4. Gyoukou

Gyoukou is owned by the Japan Agency for Marine-Earth Science and Technology. It's the successor to a supercomputer called Earth Simulator, and first entered the Top500 list in 69th place in June 2017. 

Since then manufacturer ExaScaler has added millions more accelerator cores, and by November 2017 it had climbed to 4th place.

Built around 1,250 Intel Xeon 16-core processors with an Infiniband EDR interconnect, Gyoukou derives most of its power from the 19,840,000 cores in its Pezy-SC2 accelerators.

It can deliver maximum sustained performance of 19.14 petaflops, and peak performance of 28.19 petaflops. 

Gyoukou's standout feature, though, is its power efficiency of 14.17 gigaflops/watt, almost double the efficiency of its rivals.

3. Piz Daint

Piz Daint has been climbing up the Top500 supercomputer list since November 2012, when it took 114th place with its 1,504 Intel Xeon E5-2670 processors.

Since then its owner, the Swiss National Supercomputing Center, and manufacturer Cray have performed a number of upgrades. After six months they tripled the number of processor, catapulting Piz Daint to 42nd place, and then after another six months added Nvidia K20x accelerators, boosting it to sixth place. 

In 2016, Cray switched the Xeons to a newer model, the E5-2690v3, each with 12 cores instead of the 2670's eight, and replaced Nvidia's K20x accelerators with the newer Tesla P100. A final boost to the number of processors pushed Piz Daint into 3rd place in June 2017, and it hadn't been overtaken by November.

Its 361,760 cores deliver maximum sustained performance of 19.59 petaflops, and peak performance of 25.33 petaflops. 

At 8.622 gigaflops/watt, it can't approach the power efficiency of Gyoukou, but it beats everyone else in the top 10. 

2. Tianhe-2 (MilkyWay-2)

Tianhe-2 (MilkyWay-2), at the National Super Computer Center in Guangzhou, China, was the world's fastest supercomputer from June 2013 until June 2016, when it was dethroned by another Chinese machine. As of November 2017, it's still the world's second-fastest machine. 

It uses a mix of Intel Xeon E5-2692v2 and Intel Xeon Phi 31S1P processors, for a total of 3,120,000 cores and 1 petabyte of RAM. 

Tianhe 2's maximum sustained performance is 33.86 petaflops, and theoretical peak performance is 54.90 petaflops.

Its power efficiency -- just 1.902 gigaflops per watt -- marks it as a product of its time: It was built in 2013, after all. 

1. Sunway Taihulight

For the fourth time in a row, Sunway Taihulight leads the twice-yearly Top500 list of the world's most powerful supercomputers.

Built at China's National Supercomputing Center in Wuxi, Sunway first appeared on the list in June 2016.

It has no accelerator chips, relying instead on 40,960 Sunway 26010 processors. Each has 260 cores.

Together they deliver a maximum sustained performance of 93.01 petaflops and a theoretical peak performance of 125.44 petaflops.

Power efficiency is a very respectable 6.051 gigaflops/watt.

Copyright © 2017 IDG Communications, Inc.