A new supercomputer has been launched at Moscow State University

2023: Launch of new 400 petaflops supercomputer

At the end of August 2023, Moscow State University. M.V. Lomonosov announced the launch of a new supercomputer. Its power will be used, among other things, in the search for new methods of protecting systems based on artificial intelligence (AI) technologies. The performance of the computer is 400 petaflops.

The new supercomputer will be used in research related to physics, chemistry, biology, psychology, sociology, geology, medicine, in the development of new tools based on artificial intelligence (AI), for example, algorithms for analyzing big data and other areas of science, as well as the search for new methods for protecting systems based on AI technologies.

MSU announced the launch of a new supercomputer

As stated in MSU, the new solution is based on the practices and existing developments in the field of high-performance computing of the world's leading manufacturers. The architectural basis of the solution was taken by the implemented supercomputers of the world's leading universities. The supercomputation complex includes new power supply, cooling and communication systems. According to the rector of Moscow State University named after M.V. Lomonosov Viktor Sadovnichy, the new supercomputer can enter the top three world leaders among systems of this class.

Earlier, the Moscow State University launched the Chebyshov supercomputer (60 teraflops/s), created with the participation of the SKIF-GRID program of the Union State. Then Lomonosov was installed at the university, which, after several stages of modernization, reached a peak productivity of 1.7 petaflops. The Lomonosov-2 supercomputer was launched at the university in 2012, its peak performance is estimated at 5.5 petaflops. By August 2023, it is used by more than 3 thousand users from 20 departments of Moscow State University, as well as representatives of more than 50 institutes of the Russian Academy of Sciences and more than 40 universities in Russia.^[1]

2018: MSU lawsuit against T-Platforms due to delay in contract execution

On November 1, 2018, CNews reported that the domestic company "T-platforms" had tightened the project of upgrading the supercomputer of Moscow State University named after M.V. Lomonosov (Moscow State University) - "Lomonosov 2." For this, she will have to pay Moscow State University a penalty of 45.7 million rubles. and a fine of 298.5 thousand rubles.

The amounts of financial compensation were determined by the decision of the Moscow Arbitration Court, to which Moscow State University filed a lawsuit in mid-June 2018. As follows from the documents of the proceedings, the delay in the execution of the contract between the university and T-platforms was 174 days.

In its lawsuit, Moscow State University demanded to recover from the fined contractor penalties in the amount of 90.8 million rubles. and a fine of almost 3 million rubles. However, the court found it possible to satisfy these requirements only partially.

On November 1, 2018, Moscow State University was unable to comment on the CNews completed proceedings. In "T-platforms," answering the question of CNews, whether the company intends to challenge the decision of the servants of Themis, indicated that by a court decision supercomputer there are no disagreements on the project between Moscow State University and the company. The organization also assures that Moscow State University uses the power of the supercomputer in full, and there are no technical problems with the operation of the system. The delay in the project in the company was explained by the fact that the Lomonosov 2 is based on boards with a complex topology, which involves the operation of two accelerators on one, processor which were developed and created specifically for the project, which affected the production time. "^[2]

2016: Purchase of the 2nd additional set of equipment for the supercomputer complex

Moscow State University decided to purchase additional equipment for the supercomputer complex, which is located in the Lomonosovsky university building in Moscow. For this purpose, in November 2016, the educational institution announced an auction with an initial contract price of 599.95 million rubles^[3].

Moscow State University assembles a powerful supercomputer complex in the university building "Lomonosovsky"

The customer calls the purchased equipment "The second additional set of the first stage of computing equipment for the supercomputer complex." The first such set was purchased as part of an auction held in late 2015 - early 2016. Then the only participant, T-Platforms, became the winner, and the cost of the set amounted to 395 million rubles.

In the new purchase, the university needs computing equipment consisting of 4 main modules - these are 2 sets of computing nodes, as well as control equipment and data transmission networks.

The main characteristics of the required devices are as follows:

• Total peak performance of computing nodes - at least 1.6 PFlop/s
• Total peak performance of installed computation accelerators - at least 1.5 PFlop/s
• Total peak performance of installed CPUs - at least 7.6 TFlop/s
• The total number of central processing units installed is at least 160

All supplied hardware must be guaranteed for 5 years. Additionally, within 2 years, the supplier will be obliged to perform maintenance of the installed systems.

Among other requirements, it is worth noting the need to install the supplied devices in the customer's hardware cabinet of the A7100 A-Class system rack T-Platforms, as well as demonstrate the health of the equipment on the High Performance Linpack test.

Delivery is divided into 2 stages. By April 30, 2017, the customer must receive a calculation accelerator board, by November 30, 2017 - all other equipment.

2015: Purchase of the 1st additional set of equipment for the supercomputer complex

In October 2015, Moscow State University announced an auction for the purchase of the 1st additional set of the first stage of computing equipment for the supercomputer complex. The initial value of the contract was 395 million rubles^[4]..

One application from the T-Platforms company was submitted for the auction, and as a result, an agreement was concluded with it. The value of the contract remained unchanged.

Under the terms of the agreement, the customer required additional equipment for the supercomputer complex with the following characteristics:

• Total peak performance of computing nodes - at least 375 TFlop/s
• Total peak performance of installed computation accelerators - not less than 274 TFlop/s
• Total peak performance of installed CPUs - at least 40.9 TFlop/s
• The total number of central processing units installed is at least 192

All equipment consisted of 5 main modules - this is a set of hardware cabinet equipment for installing computing nodes, a set of computing nodes, as well as 3 sets of data network equipment.

2014: New mega-powerful supercomputer of Moscow State University

On November 18, 2014, Russian supercomputer manufacturer T-Platforms announced that it was completing testing and commissioning of a new high-performance computing system for Moscow State University with a peak performance of 2.57 Pflops.

The new supercomputer became the fifth and most powerful in the university's "collection" of supercomputers. The most powerful computing system at Moscow State University before it was the Lomonosov supercomputer with a peak performance of 1.7 Pflops. As of autumn 2014, it also occupies the first line of the supercomputer rating of the Top 50 of Russia and the CIS. Three of the systems already existing at Moscow State University also supplied T-Platforms.

At the time of creation, the new Moscow State University supercomputer is also the most powerful in Russia of the systems about which there is public information. It is known that in addition to Moscow State University , Rosatom also has a petaflops performance supercomputer, but this system is not present in the Top 50 rating, and data on its total performance as of 2014 are not disclosed.

In December 2014, the computer complex with a total peak capacity of 1.5 Pflops plans to launch Rostec represented by the Solnechnogorsk Instrument Plant (SPZ). The main goal of creating the latter is to provide industrial enterprises, primarily defense industry enterprises, with high-performance computing resources.

It is worth noting that in October 2012, Moscow State University ^[5] at that time they were completing the creation of infrastructure to bring the capacity of Lomonosov to 10 Pflops, and after 1.5-2 years they planned to reach 21-25 Pflops with support from states outside.

The new, most powerful, supercomputer of Moscow State University will be located in one of the new blocks of the university

Project progress

The auction for the implementation of work on the project of Moscow State University was announced at the end of August 2014. Its initial price was 800 million rubles, and T-Platforms were the only company to submit an application for participation in the tender. According to the project TA, the deadline for completion of the work is mid-June 2015.

T-Platforms told TAdviser that the contract for the project was signed with Moscow State University at the end of October. At the same time, already in November, the new supercomputer was in the world ranking of the Top 500 most powerful computers, taking 22nd place in it.

Supercomputers in the Top 500 are ranked by the performance they achieved on the Linpack test. Under the terms of the rating, acceptance of applications for participation in it is completed about a month before the publication of the list. Thus, in order to get into the rating according to all the rules, the new Moscow State University supercomputer should have been assembled by about mid-October for the opportunity to remove the Linpack results from it.

Representatives of T-Platforms regarding this told TAdviser that "the conclusion of a contract with such a large customer as Moscow State University is always a long and difficult process, going through many instances, while the deadlines still need to be maintained, even intermediate ones - for example, if it is necessary to conduct preliminary testing of the system for submission to the Tor500 list."

"Of course, not all work is completed, the installation is not completed, and under the terms of the contract we really have the right to hand over the fully configured system only in June next year. However, most of the equipment has already been purchased in advance, and the system was assembled sufficiently to remove the first results of the Linpack test, - said representatives of T-Platforms. "Now installation, commissioning, configuration and testing will continue in a calm mode."

Purpose of the system

T-Platforms says that the number of tasks in the constantly growing queue for the Chebyshev and Lomonosov supercomputers installed at Moscow State University is in the hundreds. However, the power of the new multi-platform system is planned to be used, in many respects, for the largest and most resource-intensive calculations that can use hundreds of processors at the same time.

According to company representatives, by the time the new access system is launched, "large-scale research projects in the field of medicine (for example, modeling molecular processes in a living cell), engineering calculations (among them are the most complex problems of vortex aerodynamics and analysis of combustion processes), as well as astrophysics, materials science, power and many other areas."

Moscow State University expects that the commissioning of a new supercomputer, first of all, will allow us to start creating and processing more complex computer models than was possible before, as well as significantly unload Lomonosov and reduce the waiting time for tasks in line.

Supercomputer filling

According to the TA for the project, the equipment of the new computing complex of Moscow State University is located in three machine rooms with an area of ​ ​ 153, 83 and 51 square meters. m. It includes 5 compute racks with 1280 nodes based on 14-core Intel Xeon E5-2697 v3 processors and NVIDIA Tesla K40 accelerators, with more than 80TB total RAM. The system uses two independent Ethernet control networks and two InfiniBand FDR networks.

Each rack supercomputer consumes about 130 kW and is cooled with 45 ° C water, which is supplied to radiators closely adjacent to the computer boards. The water cooling Kontur is built using drop-free connectors, and all hydraulic and electrical components are spaced in space and protected.

The supercomputer's two-tier storage system includes a 400TB-sized calculation data store based on a parallel Lustre file system and separate access servers for more efficient handling of download files and calculation results.

2011: "T-Platforms" will design a supercomputer with a capacity of 10 Pflops

T-Platforms, an international supercomputer developer and provider of a full range of HPC solutions and services, announced in December 2011 the conclusion of a contract for the design of a computing cluster with a peak performance of 10 Pflops (1015 floating point operations per second) commissioned by Moscow State University named after M.V. Lomonosov. Under the terms of the agreement, the company will develop a computing complex that includes both a supercomputer and a complex of engineering infrastructure that ensures uninterrupted operation.

Today, Lomonosov Moscow State University has two computing complexes of T-Platforms. The Moscow State University Chebyshev supercomputer was built in 2008. At the time of launch, it occupied the 36th line of the Tor500 list of the most powerful computing clusters in the world and was the seventh most powerful among supercomputers used in the world education system. After commissioning, the resources of this cluster began to be used in full and even then their shortage became obvious.

As the volume of current scientific problems and tasks increased, there was an objective need to increase computing power. The university management decided on the need to acquire a new, more powerful supercomputer complex, which would provide high-performance resources for more scientific teams and allow more research. Such a cluster was the Lomonosov supercomputer developed and built by T-Platforms. Today, the performance of this cluster reaches a record 1.3 Pflops for Russia and the CIS, which allows it to maintain leadership in the ranking of Top50 - the most powerful supercomputers in Russia and the CIS, as well as occupying the 13th line of the Top500 list for a long time.

Currently, on the basis of computational complexes of Moscow State University, fundamental scientific and practical research is being carried out in aerospace, atomic, biomedical, oil and gas and many other scientific fields. However, due to the constant increase in the complexity of these studies, the existing computing power is not enough to conduct them on time. In this regard, it was decided to develop a fundamentally new computing cluster with a capacity of 10 Pflops, which has a sufficient margin of performance to solve the most resource-intensive problems.

have been cooperating with Lomonosov Moscow State University for many years, and without exaggeration we can call him a strategic partner of our company," says Vsevolod Opanasenko, General Director of T-Platforms. We consider the victory in the competition for the right to design a fundamentally new computing complex with a record performance for our country as additional confirmation of the highest expertise of our engineers and developers, expressed by the largest Russian scientific center. The creation of this system will mark a new stage in the development of the domestic supercomputing industry and will provide fundamentally new opportunities for the development of Russian science. "

"In Russia, the potential of supercomputers, which are combined with high-speed data transmission channels, should be fully involved. With their help, in the five-year perspective, it will be possible to design the latest aircraft and spacecraft, cars and nuclear reactors. After all, a complex technique that has not passed supercomputer modeling, as they say, is not put into numbers, in a few years it will simply not be in demand by the market. " (From the message of the President of the Russian Federation Dmitry Medvedev to the Federal Assembly of November 12, 2009

Notes