MOSCOW (Sputnik) – The Neural Networks and Deep Learning Lab of the Moscow Institute of Physics and Technology (MIPT) has launched its first NVIDIA DGX-1 supercomputer designed for deep learning tasks, local media reported Wednesday.
“The computing power is fundamentally important for deep learning, the more powerful hardware we have, the more complex neural network architecture we will be able to work with. The complexity of the model often allows us to make a revolutionary breakthrough in solving practical problems …. Good equipment will allow us to solve practical problems, which we could not even undertake without it,”
MIPT’s Neural Networks and Deep Learning Lab head Mikhail Burtsev said, as quoted by Vesti.ru portal.
The power of the supercomputer will be mainly used in the work on the iPavlov conversational intelligence project, which the lab implements jointly with Sberbank in the framework of the National Technological Initiative.
Physicists are engaged in the development of “conversational” computer intelligence, which will be able to conduct a meaningful dialogue with a human. The algorithm will not only be able to answer the interlocutor’s questions, but also request information necessary to solve the goal set in the dialogue. For this, the neural network will be trained on large arrays of documents and text records of dialogues between people.
Nvidia DGX-1 is a server with a processing power of 170 teraFLOPS aimed at accelerating deep learning applications. It is based on new Volta architecture, features eight Tesla V100 graphics processing units (GPUs) and retails at $149,000.
The statements, views and opinions expressed in this column are solely those of the author and do not necessarily represent those of The Duran.