The US supercomputer "apex" once again won the championship, and China continued to expand its lead in terms of quantity. The list shows that there are a total of 228 supercomputers in China on the list, and the number of them on the list has been ranked first, an increase of 9 units compared with the list released half a year ago. The United States ranked second with 117 units, followed by Japan, France, and Germany.
Further narrowing the gap in total computing power with U.S. Supercomputing
The new issue of the top 500 global supercomputers list was released in Denver, USA. Compared with the top ten rankings six months ago, the top ten rankings of the latest list have not changed. It is also by Lawrence Liver, a subsidiary of the US Department of Energy. "Mountain Ridge" developed by Moore National Laboratory, "Shenwei·Light of Taihu Lake" and "Tianhe No. 2" ranked second, third, and fourth. The winning "apex" was developed by the Oak Ridge National Laboratory under the U.S. Department of Energy. The floating-point operation speed reached 1.486 billion times per second, the same speed as six months ago.
The list shows that Lenovo, Sugon and Inspur rank among the top three global supercomputer manufacturers. From the perspective of total computing power, the United States accounted for 37.1% of supercomputing, and China accounted for 32.3%; half a year ago, the United States accounted for 38.4% of supercomputing and China’s The gap in the United States is further narrowing.
In fact, the pursuit of China and the United States in the field of supercomputing has been going on for several years. Before that, China’s "Shenwei·Light of Taihu Lake" had won the championship many times, and the US "Crest" was the first time. After reaching the summit, he won the championship four times in a row, but the computing speed advantage of "Top" and "Shenwei·Taihu Light" has not been further expanded.
However, in terms of the chips used in supercomputers, China is still Great disadvantage. The top two "peaks" and "ridges" on the list all use IBM's Power9 central processing unit, and from the perspective of market share, Intel is still the "leading" in the field of central processing units. The latest In the first phase of the supercomputing 500 list, about 94% of supercomputers use Intel "Xeon" or "Xeon Phi" processors.
Although the chip used by "Shenwei·Taihuzhiguang" is a "Shenwei" chip independently developed by China, more supercomputers developed by Chinese companies still use Intel chips and want to change this phenomenon , More effort is needed.
In the E-level supercomputing era, who can be the king?
As mentioned earlier, the "vertex" floating point calculation speed that won the championship reached 1.486 billion times per second, and in the future, E-level supercomputing will become a new battlefield in the field of supercomputing. E-level supercomputing is tens of exascale supercomputing, from the vertex of 1.486 billion floating-point operations to tens of exascale, there is huge room for development in the middle. China and the United States also attach great importance to E-level supercomputing, and The layout has already begun.
A demand plan published by the United States stated that it will invest 1.8 billion US dollars to develop at least two new exascale supercomputers in order to seek the United States' leadership in the field of high-performance computing. But in fact, according to the contents of the manual, the US Department of Energy plans to develop a total of 3 exascale supercomputers, and the third is a "potential system". The final development depends on the situation.
And this is the second time the U.S. Department of Energy has decided to invest heavily in development The exascale supercomputing system is now available. Its first exascale supercomputing A21 is currently in the development stage and will be deployed at the Argonne National Laboratory and will be put into use in 2021. The two supercomputers planned to be developed in April last year will be deployed in Oak Ridge National Laboratory and Lawrence Livermore National Laboratory in 2022 and 2023 to meet the advanced scientific computing research project of the Office of Science of the Department of Energy. The needs of the Nuclear Safety Administration’s advanced simulation and calculation projects.
And my country launched the E-class computer research and development plan, "Tianhe III" is part of the plan. It is reported thatThe overall plan is divided into two phases to start. The first phase’s task is to study the "E-class computer key technology" and develop three E-class prototypes; the second phase is to specifically develop E-class computers.
There is news from the National Supercomputing Center in Tianjin that my country’s self-developed new generation of exascale supercomputer "Tianhe III" E-class prototype has completed the development and deployment, and successfully passed the sub-item acceptance . Not only that, "Tianhe-3" also uses three domestic independent high-performance computing and communication chips, namely "Michuang" many-core processors, high-speed interconnect controllers and interconnect interface controllers. In addition, it includes four types of computing, storage and service nodes, hardware subsystems such as computing processing, high-speed interconnection, parallel storage, and service processing, as well as software subsystems such as system operation, parallel development, application support, and comprehensive management. Independent design.
According to the plan, next year my country will build a completely autonomous "Tianhe-3" E-class supercomputer, which can increase its computing power by 200 times compared with the "Tianhe-1".
Meng Xiangfei, director of the Application Research and Development Department of the National Supercomputing Tianjin Center, said that they will rely on the "Tianhe III" to build a high-performance computing service platform that deeply integrates supercomputing and cloud computing, big data and artificial intelligence. A series of ultra-large-scale calculations and simulations such as long-term high-resolution climate and weather forecasts, large-scale aerospace numerical wind tunnels, brain sciences and genetic engineering.