Sberbank presents Russia’s fastest supercomputer

Nov 08, 2019
  • Supercomputer specifically designed for AI purposes
  • Supercomputer to bring AI model training to new level
  • SberCloud to make supercomputer resources available to fullest extent for businesses, state-owned companies, and researchers 

 

November 8, 2019, Moscow — Sberbank and SberCloud have unveiled Russia’s fastest supercomputer, the Christofari, named after Nikolai Christofari, the first customer of the nation’s first savings bank.

 

The supercomputer was presented by Herman Gref, CEO and Chairman of the Executive Board of Sberbank, and David Rafalovsky, CTO of Sberbank Group, Executive Vice President and Head of the Technology Block of Sberbank, at the International AI Journey conference on AI development and use.

 

The computational capacity of the Christofari supercomputer will be made available to users of SberCloud, Sberbank’s cloud solution, making the design and use of artificial intelligence algorithms as fast and convenient as possible.

 

The architecture and computing power of the Christofari supercomputer let stakeholders train models based on deep neural networks in record time. The use of supercomputer resources will improve the efficiency of problem solving in a wide range of areas, like natural language processing, computer vision, automated decision-making, risk assessment and management, fraud detection, predictive analytics, creating voice assistants and chat bots, etc. The supercomputer can be used by research, commercial, and government organizations operating in various sectors of the economy, such as oil and gas, banking, telecommunications, retail, chemical, healthcare and others.

 

David Rafalovsky, Executive Vice President, CTO of Sberbank Group, Head of the Technology Block:

 

“Businesses, government, and scientific organizations need high performance computing powers of a supercomputer to work with AI algorithms. The cloud makes it available for companies of any size, so even a small startup can get an enormous computing capacity, just like a major corporation, with no need to create or grow its own IT infrastructure, or employ specialists to maintain it. The second key value is the speed of model learning and high performance computing, which is unprecedented for Russia. This cuts time to market for products and services, while also speeding up the accomplishment of tasks for businesses and science, as some machine learning tasks cannot be completed without large computing resources like that. It means that the presentation of the supercomputer on the Russian IT market will unlock the creation of completely new AI solutions and technologies.”

Sberbank and SberCloud together with NVidia created the supercomputer based on NVIDIA DGX-2 high performance nodes featuring Tesla V100 computing accelerators. The performance of the supercomputer under LINPACK tests reached 6.7 PFLOPs. The NVIDIA DGX-2 is the world’s highest performing system to tackle the most difficult AI and HPC tasks. The special-purpose computer to speed up new types of neural networks coupled with a scalable architecture helps to efficiently work with big and complicated models. This accelerates the learning process of neural networks several-fold.

The capacity of the supercomputer will be made available to clients starting December 12, 2019.