MIPT installed the supercomputer for researches in the field of AI

On September 1, 2017 the Nvidia company announced installation in MIPT of the supercomputer for researches in the field of artificial intelligence.

Hardware

From Nvidia the new generation of graphic processors which provide data processing rate in problems of artificial intelligence, x86 of architecture, comparable with 250 servers, is the cornerstone of the DGX-1 supercomputer. According to the company, it is the first supercomputer designed especially for training of artificial neural networks. The Nvidia DGX-1 platform is equipped with everything to necessary equipment rooms and the software for problems of deep learning, tool kit of development and supports popular analytical applications with support of GPU.

The last moments of installation of computing DGX-1 node in a server rack

Computing power is essentially important for deep learning. What more powerful iron is at our disposal, we will be able to work as that with more difficult neural network architecture. The complexity of model often allows to make revolutionary jump in the solution of practical tasks. So, for example, the current revolution in computer vision and speech recognition is connected, including, with growth of computing opportunities. The good equipment will allow to solve practical problems which without it we could not even undertake — the head of the laboratory of the neural systems and deep learning of MIPT Mikhail Burtsev tells.

Appointment

Capacities of the supercomputer will be generally used in work on the project on creation of colloquial artificial intelligence of iPavlov which the laboratory implements within the National technology initiative jointly with Sberbank. Researchers from MIPT are engaged in development of "colloquial" machine intelligence which will be capable to conduct meaningful dialogue with the person. The algorithm will be able not only to answer questions of the interlocutor, but also to request information necessary to solve the object set in dialog. For this purpose it the neural network will "study" at big document files and text records of dialogs between people.

Nvidia DGX-1 is very powerful platform. It is used, for example, by one of the leading commands on artificial intelligence in the world — OpenAI. Capacities of a system are enough to enter top-50 the most high-performance computers of Russia. We will use DGX for experiments with difficult models of dialogue agents — Burtsev added.

In the modern world the artificial intelligence is used in the most different areas, beginning from problems of recognition and speech synthesis and finishing with creation of the robotic systems, solutions for the financial sector and health care. Considerably reducing time for creation and training big and difficult neural networks, DGX-1 system allows researchers to create new classes of the smart applications and machines capable to study, to see and perceive the world as the person — Anton Dzhorayev, the leading specialist according to solutions of Nvidia for the AI systems said. — We are glad that technology innovations of Nvidia will allow researchers of Physics and Technology faculty to accelerate the current projects and to begin to work with tasks which solution was considered as impossible earlier.

In addition to Mikhail Burtsev's laboratory, other scientific personnel of institute which also researches, using artificial neural networks for development of new drugs also will be able to work on supercomputers.