RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

Christofari Supercomputer

Product
The name of the base system (platform): Nvidia Tesla
Developers: Sberbank, Cloud Technology (SberCloud SberCloud)
Last Release Date: 2021/11/11
Technology: Supercomputer

Content

Main article: Supercomputers

"Christopharie" is a supercomputer created specifically for working with would-be artificial intelligence. Included in the AI Cloud product line.

2021: Christofari Neo Presentation

Sber November 11, 2021 announced TAdviser that, together with its subsidiary, the SberCloud provider Cloud high-performance Russian supercomputer Christofari Neo was introduced. It was presented by STO Sberbank Group, executive vice president. David Rafalovsky

Sber's new supercomputer Christofari Neo is almost twice as powerful as its predecessor

The effective performance of Christofari Neo in double precision is almost 12 petaflops (11.95 PFLOPs). Christofari Neo is based on the technologies of Nvidia, the world representative in the field of AI. The computing heart of the system was the Nvidia A100 graphics processors with 80 GB of memory. The A100 tensor cores provide high performance in AI tasks, and 80 GB of memory will allow you to work with large AI models and data arrays.

Christofari Neo is great for learning and inspiring transformer models with the most up-to-date architecture, which are taught on large datasets. The first test access to Christofari Neo was received by the SberDevices and Sber AI teams. Thanks to ML Space, the SberDevices team was able to seamlessly switch between supercomputers and optimize the cost and speed of training transformer models ruGPT-3 & family, available for further training and testing in SberCloud.

Шаблон:Quote 'author = said David Rafalovsky, STO Sberbank Group, Executive Vice President.

2020: The ability to use Christofari to work with any category of personal data

On March 31, 2020 Sberbank , he announced that cloudy the platform AI Cloud designed to solve problems in the field artificial intelligence had received a certificate of compliance with security requirements for information systems (personal data HID) with the provision of the second level. This security of personal data gives platform customers the opportunity to use supercomputer Christophari to work with any -, biometric special, public and other - categories of personal data and create services based on AI -algorithms using advanced medical diagnostics, voice and facial recognition other biometrics technologies.

Thus, Christofari has become a supercomputer on which you can process personal data in accordance with the requirements of Russian regulators.

File:Aquote1.png
So far, Russian companies and organizations have not had the opportunity to use supercomputers in processing personal data. Now, business and government agencies have become available the computing power of Christofari, a productive supercomputer in Russia. AI Cloud services will not only quickly train software models of artificial intelligence, but also use data sets containing personal data for training. At the same time, we guarantee customers complete confidentiality and protection of such data. For the end user, this means that he will be able to quickly obtain services from companies and organizations working with personal data in the commercial and public sectors,
File:Aquote2.png

The effective performance of Christofari for March 2020 is 6.7 PFLOPs (according to the LINPACK test). Peak performance - 8.8 PFLOPs.

2019

Commercial launch

On December 16, 2019, Sberbank and SberCloud announced the launch of the Kristofari supercomputer into commercial operation and the availability of its capacities for customers - legal entities. You can order the service through the SberCloud website.

SberCloud began to provide services based on the Christofari supercomputer (photo - TAdviser)

In the November ranking of the top 500 supercomputers in the world, Christophari, which has a real performance of 6.7 Pflops, takes 29th place and is the most powerful computer system submitted to the list from Russia.

Christofari was designed to solve a certain application problem - training models using artificial intelligence, the SberCloud explains.

File:Aquote1.png
We see our mission in making supercomputing and developing artificial intelligence solutions available to any company, regardless of its size, industry and territorial location. Previously, only the largest corporations could use such technologies, and today, with the help of Kristofari, any start-up startup or small business company, "said David Rafalovsky, STO Sberbank, executive vice president, head of the Technology block.
File:Aquote2.png

David Rafalovsky at the presentation of the supercomputer (photo - TAdviser)

As part of the SaaS model service, customers are provided not only with computing power for parallel computing and a set of necessary system software, but also with application software for this specific task - Jupyter Notebook, SberCloud Managing Director Mikhail Lobotsky told TAdviser.

File:Aquote1.png
The end user of our service is data scientist. We provide him with a ready-made workplace with the usual interface - Jupyter Notebook. This is the most often used tool for the work of such specialists, "said the managing director of SberCloud in a conversation with TAdviser.
File:Aquote2.png

This is SberCloud called one of the key differences between its service and competing ones, which, however, are not so numerous in the Russian market. Resources for high-performance computing over the Internet, for example, are provided by Moscow State University, a supercomputer modeling center based on the Solnechnogorsk Instrument Plant (part of Rostec), T-Services (a subsidiary of T-Platforms).

Another key difference in the SberCloud is the payment model: money is charged only directly during the training of the model. And the loading of data, their markup, preparation is not charged, explained TAdviser in the company.

The supercomputer occupies one engine room in the Sberbank data center in Skolkovo (photo - TAdviser)

The first commercial user of Kristofari was Sberbank itself, and on general terms developed for all customers, they say in the SberCloud.

Mikhail Lobotsky told TAdviser that at the time of the announcement of the commercial availability of concluded contracts with other customers, however, the SberCloud sees interest in the service and is testing with a number of interested companies. Their names are not disclosed in the SberCloud, but they are said to be represented by different industries.

File:Aquote1.png
Due to the fact that there are more and more tasks where artificial intelligence is used, there are not enough systems previously created for this. Therefore, we expect high demand for our service, "said Mikhail Lobotsky in a conversation with TAdviser.
File:Aquote2.png

AI-based model training is in demand in the financial sector, says SberCloud Managing Director. According to him, colleagues from other banks also turn to them with requests.

Racks with equipment of the supercomputer "Christopharie" (photo - TAdviser)

Also, Lobotsky says, there is demand in the oil and gas sector, in particular, in geological exploration for modeling deposits. In this case, we are talking about demand from oilfield service companies providing services to oil and gas producing companies, the managing director of SberCloud said in a conversation with TAdviser. Training of models in medicine is in demand, as well as in retail - to predict demand.

One of the areas where AI-based model training is required is the training of voice assistants for call centers, the representative of SberCloud gave an example. As a marketing move to Kristofari, using real voice samples, the voice of the head of Sberbank, German Gref, was synthesized. The set of data on which you can teach models in this vein is becoming smaller and smaller, said Mikhail Lobotsky.

System for monitoring supercomputer performance on the way to the machine hall (photo - TAdviser)

At the time of the start of commercial operation, in order to attract more customers, SberCloud launched a campaign: "100 rubles for 100 minutes of all the power of the Christophary supercomputer to train models," customers can get 100 minutes of work on the amount of data up to 10 gigabytes for a symbolic fee. The promotion will last until June 12, 2020.

The Kristofari supercomputer is located in the Sberbank data center in Skolkovo, it occupies one engine room. One of the features of this data center, its developers call it 85% cooled using free cooling technology - due to air from the environment. Air is taken in and subjected to two-stage cleaning, after which it is directed to the halls. The system operates at an ambient temperature below 20 ° C. If necessary, it is cooled by classic chillers.

Christofari supercomputer can become the core for creating fully automatic production

Opportunities and ways to use artificial intelligence in production companies will be the topic of discussion within the IT in Industry section at TAdviser SummIT 2019. The director of business development in the public sector of the company SberCloud Kayumov Shamil will tell the forum participants how enterprises can modernize production and avoid high costs by introducing modern IT and why working in the cloud is more profitable for the company than maintaining their own infrastructure.

Digitalization of the production sector in Russia by 2025 will be able to annually increase the country's GDP by the amount of 1.3 trillion rubles to 4.1 trillion rubles. According to a study by McKinsey Global Institute, the use of digital technologies will increase enterprise productivity by 45-55%, as well as reduce the time to market by 20-50%.

According to Shamil Kayumov of SberCloud, the high degree of depreciation of fixed assets is becoming an increasingly server problem for enterprises and they require the modernization of industrial industries, while all modern methods of production management are based on IT.

Production control systems - ERP systems and MES systems - can operate from the cloud. The company SberCloud provides customers with cloud infrastructure and, thus, will ensure the modernization of production, while the capital costs will be significantly less than if the enterprise used its IT infrastructure.

The Christofari supercomputer, which Sberbank and SberCloud presented on November 8 at the AI Journey international exhibition, can be the core for creating fully automatic production, providing a transition to the so-called "deserted" technologies.

File:Aquote1.png
SberCloud (Technology LLC Cloudy) is a cloud platform of the Sberbank Group that provides services based on the IT architecture of the largest bank in Russia, the CIS and Eastern Europe. Infrastructure, information technology platforms and services are SberCloud the basis of Sberbank Group's digital ecosystem, and are also provided to external clients - companies and state organizations.
File:Aquote2.png

Christofari entered the top 30 most powerful supercomputers in the world

In November 2019, an updated rating of the most powerful supercomputers in the world was published. Christofari took 29th place in it. More details here.

Supercomputer Announcement

On November 8, 2019, Sberbank and its cloud provider SberCloud introduced a supercomputer, which developers call the most powerful in Russia. It is planned to use its resources to create and develop internal services of the Sberbank ecosystem based on artificial intelligence, as well as sell as a service to users of the SberCloud cloud service.

Sberbank says that the performance of the supercomputer in Linpack tests (the most popular benchmarks for measuring the power of high-performance systems) reached 6.7 Pflops. For comparison, the real performance of the most powerful supercomputer in Russia in the ranking of the top 50, compiled by Moscow State University in September, is about 2.6 Pflops. This is the Lomonosov-2 supercomputer installed at Moscow State University, the developer of which is T-Platforms.

Sberbank's supercomputer, named Christofari in honor of the first client of Russian savings banks Nikolai Christofari, was created in partnership with Nvidia on the basis of high-performance Nvidia DGX-2 nodes equipped with Tesla V100 computing accelerators.

Presenting the supercomputer, Sberbank CTO, executive vice president, head of the Technology block David Rafalovsky noted with excitement that Christofari is the "main ingredient" of Sberbank's AI cloud.

David Rafalovsky introduced a new Sberbank product "(photo - itweek.ru)"
File:Aquote1.png
One of our most important goals is to make AI available to everyone, "said Rafalovsky.
File:Aquote2.png

The computer is designed to solve problems in the development and use of artificial intelligence algorithms in order to accelerate their implementation several times. For example, it can quickly train models based on deep neural networks.

The machine can be used in a wide range of areas, such as natural language processing, computer vision, automated decision making, risk assessment and management, fraud detection, predictive analytics, the creation of voice assistants and chat bots, as well as in other areas.

One of the first users of the supercomputer is a partner and company within Sberbank itself - the VisionLabs company, which uses a computing system for parallel processing of many video streams. Another user is S7 airline, where they tested a supercomputer on training a model that solves the problem of predicting the demand for airline tickets.

As of November, Sberbank itself is already using a supercomputer for several banking business processes, for example, when recognizing and synthesizing speech when automated analysis of customer calls to the bank's call center, for an automated robot operator. Previously, the bank trained AI for these purposes for a month, with a supercomputer - several hours, Sberbank says.

Supercomputer services will become available to customers from November 12, Rafalovsky announced. The cost of services will depend on the amount of capacity ordered. Rafalovsky announced the price of using the entire supercomputer - 5.75 thousand rubles per minute. This is "exceptionally cheap," Sberbank's CTO said.

He also voiced a promotional action: for 100 rubles per minute SberCloud offers 100 minutes of all Christofari power. This proposal at Sberbank is considered convenient for those who want to try how it works.

At the same time, Sberbank is ready to provide access to Christofari resources to universities and scientific organizations for a symbolic fee or even for free at times when the computer system will not be loaded by commercial customers.

Chairman of the Board of Sberbank German Gref supplemented the presentation with a statement that if the system is heavily loaded with orders, the bank is ready to invest in the creation of another supercomputer.

I must say that Sberbank is far from the first to offer commercial customers a remote lease of supercomputer resources and related services. Of the large organizations, for example, Rosatom, which also has one of the most powerful, and maybe the most powerful supercomputer in the country, began to do this even earlier. The fact is that Rosatom does not disclose the performance indicators of its car.

The developer of T-Platforms supercomputers began to provide such services even earlier, creating a separate company for this in 2008 - T-Services. In addition to leasing computing capacity, the company also offers services in mathematical modeling of physical processes, the development of special software, etc. The revenue of T-Services at the end of 2018 amounted to about 80.5 million rubles.

In 2018, Solnechnogorsk Instrument Plant (SPZ), which is part of Rostec, commissioned a center for high-performance computing and modeling, albeit with a significant delay. It was also planned to provide remote access services to high-performance resources and related services. In Roselectronik, which includes SPZ, in July 2019, TAdviser stated that in the segment of providing supercomputer resources in Russia there is a shortage of offers.

MTS also in 2018 launched cloud services for high-speed data processing based on a supercomputer. Like the Sberbank supercomputer, it uses the Nvidia solution with Tesla V100 video cards. The company offered several charging options: with the possibility of daily use of a supercomputer from less than 1 thousand rubles per day for small companies and in subscription format from 29 thousand to 200 thousand rubles per month, depending on the required processing power for large businesses.

MTS told TAdviser that the demand for the service is steadily high, in the plans of the cloud provider# CloudMTS a further increase in the resources of the supercomputer for offering to customers.

File:Aquote1.png
Since the launch, several large customers have been connecting to the cloud supercomputer every month, using it in big data and artificial intelligence projects, "the MTS representative said in a conversation with TAdviser.
File:Aquote2.png