Not everyone understands at all why we need some expensive supercomputers. Maybe it’s a ponte and an extra waste of money? Let’s try to figure it out …
The new Top500 rating of the world’s most powerful supercomputers has been published . This time, the American machine took the first place, pushing the Chinese “colleague”. IBM launched the most powerful supercomputer Summit, which was created by the order of the National Laboratory of Oak Ridge of the US Department of Energy. The performance of its system was estimated at 200 petaflops (200 quadrillion operations per second). While China and the US are fighting for “medals”, Russian cars …
… they are content only with the place in the eighth ten. Ilya Plekhanov talks about the computational race and about the new opportunities of the largest countries of the world with their powerful machines. </ strong>
After the standard Linpack test, which fixes the performance of the supercomputer, the real value of the American Summit was 122.3 petaflops, and the performance of the former champion, the Chinese Sunway TaihuLight, was 93 petaflops. So the United States returned the palm itself in the most productive supercomputers, which China lost in 2013.
The new American Summit consists of 4 608 dual-processor IBM AC922 servers based on 22-core IBM Power9 processors, each of which has 6 NVIDIA Tesla V100 graphics processors. For internal connections, a Mellanox dual-rail EDR InfiniBand network is used with a bandwidth of 200 Gb / s per server. The supercomputer runs under the operating system Red Hat Enterprise Linux 7.4. MIT Technology Review reports that Summit weighs more than a passenger airliner, and occupies a space of 2 tennis courts.
Even before the final completion of the system, the researchers launched calculations comparing genomes, during which Summit’s performance was fixed at 1.88 exaflops. According to Thomas Zacharias, director of the Oak Ridge National Laboratory, this is the first time in history when the barrier in 1 exaflops was overcome. In the winter, China threatened to launch a computer with a performance of 1 exaflops, that is, a quintillion (10 to 18 degrees) of operations per second by the end of this year, and in the US, $ 258 million was previously allocated to create a supercomputer with a performance of more than 1 exaflops by 2021. Development, as announced in 2017, is led by six US companies – Hewlett-Packard, IBM, Intel, NVIDIA, Cray and AMD.
Sound of a working supercomputer
Symbol of the power of the country
The appearance of the next supercomputer is a kind of prestigious landmark and an indicator of what the country can do, whether it has the resources and skill, the budget from hundreds of millions to over a billion to build a supercomputer, whether there are power capacities for food, engineering solutions for heat dissipation, whether there are processors, qualified frames and so on. US Energy Secretary Rick Perry commented on Summit’s appearance:
Today’s launch of the Summit supercomputer demonstrates American leadership in the field of scientific innovation and technology development. This will have a huge impact on research in the field of energy, scientific discoveries, economic competitiveness and national security.
It is the issue of national security that can be considered a key issue, since the US is at least worried about China’s intention to become a leader in the field of artificial intelligence. This after all means the use of AI in the military sphere.
The Chinese Sunway TaihuLight, which took second place, was a breakthrough for China in creating its own computing and elemental base. The Sunway TaihuLight processors are based on the local ShenWei architecture created at the Shanghai High Performance IC Design Center in Shanghai. The system of interconnection of nodes, known as the Sunway Network, was also developed in China. In addition, Sunway TaihuLight has its own implementation of the software standard for parallel programming – the operating system SunWay Raise OS 2.0.5 based on Linux. That is, China was able to overcome dependence on American elements and imposed restrictions on the supply of high-tech products from 2015.Sunway TaihuLight
In Russia in 2009, the Lomonosov supercomputer was installed at the Moscow State University, which contains 6,654 computing nodes, more than 94,000 processor cores, and has a peak performance of 1.37 Pflops. The actual performance of the system on the Linpack test is 674 Tflops, which allowed it to take the 13th place in the Top500 list of the world’s most productive computers in June 2011.
In March 2018, MSU announced that the development and experimental operation of a new section of the Lomonosov-2 supercomputer with a peak performance of 1.8 Pflops had been completed, which in the future would allow it to approach the 5 Pflops boundary. The new section of the supercomputer is based on multi-core Intel Skylake processors and NVidia Pascal P100 graphics processors with hardware support for the high-speed NVLink interface.
In March, the staff of the Joint Institute for Nuclear Research in the Moscow region of Dubna presented a new supercomputer “Govorun”, which will be used to process data received from the future heavy ion collider NICA. The productivity of Govorun is 1 petaflops. The supercomputer is built on the basis of 72-core Intel Xeon Phi 7290 and Intel Xeon Gold 6154 processors.
Supercomputers in Russia appear, but not such a performance to get into the Top500 rating. In January of this year, Igor Chernykh, then acting as the head of the Central Supercomputer Center of the Siberian Branch of the IOMMG SB RAS, said :
Supercomputers in Russia are produced by several companies, the most famous are RSK, “T-Platforms”. However, our component base is 99% imported. One of the last places in the world for production and supply is Russia. Abroad, supercomputers are almost not supplied.
According to Oleg Gorbachev, Director of Corporate Communications of the RSC Group, the situation could be corrected by the national program:
For the development of production, we just need a national program. The USA and China have the lion’s share of these systems in the rating precisely because they have a national program, because there is a state understanding that this is necessary, and all this is outlined in the programs and reinforced in the budgets.
A supercomputer is needed in order to work with a huge amount of data and variables – with information measured in exabytes (10 to 18-byte bytes), and in iotabytes (10 to 24-byte bytes). And it’s not just about volume, but speed. For example, the calculation of meteorological conditions will be meaningless if they last longer than the onset of a particular weather phenomenon.
Another issue is that the difficulties arise not so much with the number of operations per second, but with the elaboration of those algorithms that require such speeds and so many processors. That is, technical capabilities are being improved, but they need mathematical models, software stuffing – understanding how and what to calculate.
Doctor of Physical and Mathematical Sciences, Chief Researcher of IPM. M.V. Keldysh Andrey Kuleshov explains the specifics of using supercomputer technology:
Designed for low-power machines and effectively on them, the functioning software is completely inappropriate for supercomputers, especially hybrid architectures, where together with conventional processors, for example, graphics accelerators are used. Without creating special algorithms that are designed to work on these computing architectures, the capabilities of supercomputers will not be realized.
Therefore, Russian scientists believe that Russia’s lagging behind the US and China in petaflops is not so scary. The key to effective use of machines is a high mathematical level and a class of software specialists in Russia.
What are they doing now?
For example, the Spanish supercomputer MareNostrum (the so-called Mediterranean Sea of ancient Romans-pagans) is located in the premises of the former Christian chapel of Torre Girona and is used for calculations of genetic engineering, astrophysicists, when creating new medicines and forecasting the weather.
But not all is obsessed with large-scale scientific research. Business is also actively using supercomputers. For example, Ford uses powerful machines to perform crash tests, when cars are broken against an obstacle and it is necessary to take, measure and analyze a lot of parameters. Using a supercomputer allows you to reduce the number of broken real cars and the cost of tests by a third, saving millions of dollars.
DuPont, with the help of supercomputers, synthesized a new material – non-flammable, corrosion-resistant and low-toxic. In one week, the necessary calculations were made on a supercomputer, for which about $ 5,000 was spent in general. The study “in the old manner” would cost at least $ 50 thousand and three months of work.Simulation of a crash test by a supercomputer taking into account damage to all elements of the machine
Since 1990, the partner of the Wimbledon tournament in technology – the American company IBM, which provides the organizers of the championship supercomputer Watson. With its help, social networks are analyzed to identify positive and negative moods of fans, to give commentators any comparative data from archives, to automate marketing and PR on the Internet.
The French company Total in oil exploration uses a supercomputer to analyze information about the subsoil. Simulation allows you not to resort to drilling when there is doubt about the availability of oil. In one such case, the abandonment of drilling saved the company about $ 80 million.
As for the new American supercomputer, some projects will have access to it already this year to model the subatomic behavior of materials, identify factors that cause cancer, and study the interaction of proteins and cellular systems. The main customer in 2019 will be the Ministry of Energy, which uses it for its calculations in the field of alternative energy. Summit will also be used in the application of artificial intelligence in 2019.
With the help of the Russian supercomputer, the Mekhmat group of the Moscow State University and the Institute of Applied Mathematics of the Russian Academy of Sciences obtained results on the numerical modeling of the formation and development of end vortices in supersonic regimes, seismic data were processed, for the needs of RSC Energia, calculations were made of the flow past the spacecraft during braking in the atmosphere and landing on the surface Of the Earth. Lomonosov also solved the problem of mass exchange for a prospective nuclear reactor. In April, the Physics Department of the Moscow State University investigated various types of polymerization reactions using computer simulations on the Lomonosov-2 supercomputer.
In the spring of this year, it became known that Rosatom and Roskosmos are starting to experience engines for a new Russian superheavy missile in the virtual reality as part of modeling on a supercomputer. The project began in 2014 and cost about 270 million rubles. January 30, 2018 signed an act on the acceptance of works on the development of software for virtual tests.
What can the future
There are many options. Supercomputers can be used to increase the accuracy of weather forecasts, diagnoses, to identify the connection between genes and cancers, to simulate turbulence, combustion processes, to develop new materials, new drugs, to simulate the work of new technology, to calculate deformations of solids, taking into account the processes of destruction, control over drones, modeling the use of modern weapons, for analyzing data from satellites and drones, cryptography, modeling the effects of electromagnetic radiation (for example, on the coverage of aircraft), modeling of brain work and so on.
Supercomputers will also be involved in personal profiling and the introduction of face recognition systems, monitoring of activity and social rankings in China. According to some forecasts, supercomputers will be able to accurately simulate the Earth’s weather system as a whole, after reaching a performance exceeding 1 zetaflops (1000 exaflops). The appearance of such machines is expected by 2030.
Not only is the performance of the best supercomputer important, but also the total number of machines in the country and their availability. The more data centers, the more organizations and institutions can access them, the greater the range of tasks can be resolved.
In Russia, only 2,500 specialists from 20 MSU departments, more than 100 institutes of the Russian Academy of Sciences, and more than 100 Russian universities use the capabilities of just one supercomputer complex of the Moscow University.
According to the Top500 project, China is the country with the greatest number of super-power computers for June this year: it owns 206 supercomputers from the rating (six months ago it was 201 system). At the same time, the United States reduced its share in the rating from 143 systems to 124. But it can be noted that by the total productivity of all systems in the country, the USA this time overtook China. The American total capacity is 38.2% of the total Top500 performance, and the Chinese one – 29.1%.
After China and the United States, Japan (36 systems), Great Britain (22 systems), Germany (21 systems), France (18 systems) are in the list of supercomputers in the ranking.
Russia has four supercomputers in the ranking – one more than in the previous list six months ago. “Lomonosov-2” took 72nd place, computer in Roshydromet – 172nd, “Lomonosov” – 326th position, supercomputer at the Kurchatov Institute – 458th line of the rating.
And further advanced countries will work on overcoming physical limitations of existing supercomputers. Companies will focus on the creation of technologies such as silicon photonics, neuromorphic and quantum computing, the creation of biocomputers and cellular computers. For example, IBM in 2014 in the first areas were opened research programs with a funding volume of $ 3 billion.
Accelerating the development of new materials and technologies with the help of supercomputers in different directions can ultimately give a cumulative tremendous impetus to the development of countries and their technological separation from competitors. The rapid growth of computer productivity is called even a revolution in the development of civilization.