06.02.2023, 13:00
Yandex warned about the risks of a shortage of Nvidia graphics cards for the development of AI and drones
Source: OREANDA-NEWS
OREANDA-NEWS The issue of the shortage of NVIDIA chips was discussed at a meeting with representatives of the Ministry of Finance and other interested structures, sources told Vedomosti. The results of the meeting and the proposed measures are not reported. Yandex did not comment on the situation.
The company was supposed to receive a large batch of NVIDIA graphics cards in March 2022, but after the start of the "special operation", the delivery failed — the manufacturer announced the suspension of sales in Russia. After that, the import of chips into the country became possible only by parallel import.
In the amount that Yandex requires, it is problematic to import video cards through parallel import, one of the sources claims. Now the company uses more than 1.5 thousand NVIDIA A100 GPUs in its supercomputers. They are needed to improve the speed and accuracy of search, Yandex Cloud, voice assistant, video streaming and other tasks.
The problem is serious for the company, another interlocutor noted. Yandex has invested a lot of resources in the development of AI, but cannot deploy an infrastructure for it without video cards.
The current pace of neural network development implies a constant complication of models and an increase in equipment requirements, says Smart Engines CEO Vladimir Arlazarov. The shortage of Nvidia GPUs is a serious problem for the development of artificial intelligence in Russia, notes Smart Engines CEO, PhD Vladimir Arlazarov: “Sanctions have been imposed on high-performance equipment for training neural networks, in addition, some companies, for example, Sberbank, are under US sanctions, their own comparable hardware and capacities for their Unlike China, Russia does not have production."
As part of the current paradigm of neural network development, an extensive approach is used in Russia to achieve new levels of quality and functionality, says Vladimir Arlazarov. Such an approach, according to him, provides for a constant complication of models and, accordingly, an increase in equipment requirements. “In addition, thoughtless and poorly informed use of neural networks is characteristic for solving any problems, even if they can be perfectly solved by other approaches. Therefore, it will be difficult for companies that use this approach - and among them Yandex, Sberbank, and VK - to maintain the pace of development,” he believes.
According to Vladimir Arlazarov, the solution may be the transfer of development centers to a neutral territory, purchases on gray imports, or a complete change in the paradigm of AI development, namely, the transition from an extensive approach typical of Western and Chinese schools to an intensive, characteristic of the Soviet school path of development. “Most likely, the problem will be solved by a combination of the first and second ways. Choosing an intensive path of AI development will require shifting the focus to fundamental science and the Russian Academy of Sciences as the main actor, although this path is the most profitable for Russia, it is unlikely now,” concludes Vladimir Arlazarov.
The company was supposed to receive a large batch of NVIDIA graphics cards in March 2022, but after the start of the "special operation", the delivery failed — the manufacturer announced the suspension of sales in Russia. After that, the import of chips into the country became possible only by parallel import.
In the amount that Yandex requires, it is problematic to import video cards through parallel import, one of the sources claims. Now the company uses more than 1.5 thousand NVIDIA A100 GPUs in its supercomputers. They are needed to improve the speed and accuracy of search, Yandex Cloud, voice assistant, video streaming and other tasks.
The problem is serious for the company, another interlocutor noted. Yandex has invested a lot of resources in the development of AI, but cannot deploy an infrastructure for it without video cards.
The current pace of neural network development implies a constant complication of models and an increase in equipment requirements, says Smart Engines CEO Vladimir Arlazarov. The shortage of Nvidia GPUs is a serious problem for the development of artificial intelligence in Russia, notes Smart Engines CEO, PhD Vladimir Arlazarov: “Sanctions have been imposed on high-performance equipment for training neural networks, in addition, some companies, for example, Sberbank, are under US sanctions, their own comparable hardware and capacities for their Unlike China, Russia does not have production."
As part of the current paradigm of neural network development, an extensive approach is used in Russia to achieve new levels of quality and functionality, says Vladimir Arlazarov. Such an approach, according to him, provides for a constant complication of models and, accordingly, an increase in equipment requirements. “In addition, thoughtless and poorly informed use of neural networks is characteristic for solving any problems, even if they can be perfectly solved by other approaches. Therefore, it will be difficult for companies that use this approach - and among them Yandex, Sberbank, and VK - to maintain the pace of development,” he believes.
According to Vladimir Arlazarov, the solution may be the transfer of development centers to a neutral territory, purchases on gray imports, or a complete change in the paradigm of AI development, namely, the transition from an extensive approach typical of Western and Chinese schools to an intensive, characteristic of the Soviet school path of development. “Most likely, the problem will be solved by a combination of the first and second ways. Choosing an intensive path of AI development will require shifting the focus to fundamental science and the Russian Academy of Sciences as the main actor, although this path is the most profitable for Russia, it is unlikely now,” concludes Vladimir Arlazarov.
Комментарии