10, computer
KComputer, built by Fujitsu for the Institute of Computer Science of Riken, Japan, was the fastest computer in the world in June 20 1 1. By June of 20 17, 1 1 year, it was successfully reduced to the position of 10 in the global supercomputer rankings. KComputer ranks the highest among Sparc-based computers in the list, using 88, 128 SPARC64VIIIfx eight nuclear processors, with a running speed of 2GHz and 705,024 cores. It has the maximum continuous performance of 10.5 1petaflops and the peak performance of 1 1.28petaflops, but it has nothing to be proud of: its performance per watt is only 0.83gigaflops, which is the least efficient machine in the top ten.
9. Oak forest PACS
The ninth Oakforest-PACS is another machine built by Fujitsu, which was built for Japan United Advanced High Performance Computing Center. Based on Fujitsu PrimergyCX 1640 cluster and Intel Full Path Interconnection, Oakforest-PACS entered the sixth place in 20 1 16, but slipped to the ninth place one year later. Its 8 178 Intel Xeon processor * * has 556, 104 cores. The highest performance can reach 13.55petaflops, and the peak performance can reach 24.9 1petaflops. Its power consumption efficiency is 4.986 gigacycles per watt.
8. Currie
Corey was built by Clay Company, which belongs to the US Department of Energy and is located in Lawrence Berkeley National Laboratory. Currie 20 16 1 1 entered the top 500 list, one place higher than Japan's Oakforest-PACS, but it also declined all the way. 20 17 1 1 ranked eighth. Another similarity with Oakforest-PACS is that Intel's XeonPhi processor is also used. There are 9 152 cores, a total of 622336 cores. The maximum sustained performance is 14.0 1petaflops, and the peak performance is 27.88petaflops. The energy efficiency of Corey is lower than that of Oakforest-PACS in Japan, and it is 3.558gigaflops per watt, which is not very impressive.
7. Sany
Trinity, which was built by Cray Company for the Los Alamos National Laboratory of the US Department of Energy, was upgraded in the summer of 20 17, pushing it behind the top 500 list. 20 15, 1 1, ranked sixth in June, equipped with18,816 Intel Xeon 5 16 core processors, but the ranking declined steadily and once fell to tenth place. On October 20 17, 165438+ 10, the upgraded Trinity was equipped with 9984 IntelXeonPhi68 core processors, which climbed to the seventh place, with * * * 979968 cores, and the maximum sustained performance and peak performance reached14 respectively. Generally speaking, its power consumption is lower and its efficiency is higher, which is 3.678 gigacycles per watt.
6. Sequoia
Sequoia was the most powerful supercomputer in the world in June 20 12, but its ranking has been declining since then. By 20 17, 1 1, it has slipped to the sixth place. Like Currie and Sany, Sequoia belongs to the US Department of Energy and is located in Lawrence Livermore National Laboratory. The difference is that Sequoia was built by IBM, using the company's 16 core PowerBQC chip and customized interconnection technology. It has 98,304 processors and 65,438+0,572,864 cores, and each core has 1G of RAM. The highest continuous performance of 17. 17petaflops and the highest performance of 20. 13petaflops are realized. In terms of energy efficiency, it can provide medium power of 2. 177 gigacycles/watt.
5. Titan
Titan defeated the original Sequoia when it entered the top 500 list at 20 12 and1,but it had slipped to fifth place before 20 1 1. It is the most powerful computer of the US Department of Energy, located in Oak Ridge National Laboratory. Like Corey and Sany, they are all produced by Cray Company, but different from the latter two, it uses AMD's 16 core Snapdragon 6274 chip and NvidiaTesla accelerator. It has a total of * * * 560,640 cores, the maximum sustained performance reaches 17.59petaflops, and the peak performance reaches 27. 1 1petaflops. The energy efficiency reaches 2. 143 gigacycles/watt.
4、Gyoukou
Gyoukou is owned by Japan Marine Geoscience and Technology Agency. It is the "successor" of a supercomputer named EarthSimulator, and entered the top 500 list for the first time in June 20 17, ranking 69th. Since then, the manufacturer ExaScaler has added millions of accelerator cores, and by 20 17 1 1, it has climbed to the fourth place. 1250 Intel Xeon 16 core processors are interconnected by InfinibandEDR. Most of Gyoukou's performance comes from19.84 million cores accelerated by Pezy-SC2 accelerator. The continuous performance of 19. 14petaflops and the peak performance of 28. 19petaflops can be realized. The outstanding feature of Gyoukou is that its efficiency is per watt 14. 17gigaflops, which is almost twice that of competitors.
3. Pitt Dane
PizDaint has been climbing on the top 500 list since June 20 1 12, when it was equipped with 1, 504 Intel Xeon processor 5-2670, occupying the position of 1 14. Since then, all its institutions-Swiss National Supercomputing Center and manufacturer Cray Company have carried out a series of upgrades. Six months later, they tripled the number of processors and upgraded PizDaint to 42nd place. Six months later, they added the NvidiaK20x accelerator, which promoted it to 6th place. In 20 16, Cray replaced Xeons with a new model E5-2690v3, each model has 12 cores instead of 8 cores of 2670, and replaced K20x accelerator with a new NvidiaTeslaP 100. In the end, the increase in the number of processors pushed PizDaint to the third place in June of 20 17, and it has not been surpassed by June of 1 1. Its 36 1 760 cores provide the maximum sustained performance of 19.59petaflops and the peak performance of 25.33petaflops, and its efficiency is 8.622gigaflops per watt.
2. Tianhe-2
Tianhe-2 belongs to China Guangzhou National Supercomputer Center. From June 20 13 to June 20 16, it was the fastest supercomputer in the world. Later, he was pushed off the altar by the light of Taihu Lake. But as of 20 17, 1 1, it is still the second fastest machine in the world. It uses Intel Xeon processor 5-2692V2 and Intel Xeon processor 3 1s 1p, with a total of 3 120000 cores and 1PB RAM. The highest continuous performance of Tianhe-2 is 33.86petaflops, the theoretical peak performance is 54.90petaflops, and the power consumption efficiency is only 0 1.902gigaflops per watt/kloc. After all, it was built in 20 13. According to the latest news, Tianhe-2 will replace the original Intel accelerator with a brand-new domestic Matrix2000 accelerator at the end of this year, officially becoming a real domestic supercomputer, and its maximum computing speed will be greatly improved, reaching nearly three times the current performance.
1, the light of Taihu Lake
Shenwei Taihu Light led the list of the top 500 supercomputers in the world for the fourth time in a row, located in Wuxi Center of National Supercomputing, and made its debut in June 20 16. It has no accelerator chip, and relies on 40,960 Shenwei 260 10 processors, each with 260 cores, providing the maximum sustained performance of 93.05438+0 petaflops and the theoretical peak performance of 125.44petaflops, and the power consumption efficiency is very impressive, with 6.05 1gigaflops per watt.