Fplus has started producing servers with neural network algorithm accelerators from the domestic microelectronics manufacturer, STC Module. Special-purpose computing devices, which are connected to the PCIe expansion slot on the motherboard, are used to implement neural networks and solve digital signal and image processing tasks. They are compatible with the Voskhod and Buran dual-socket and quad-socket servers from Fplus, which were released earlier and included in the register of the Ministry of Industry and Trade.
In his report at the plenary session during CIPR-2024, Prime Minister Mikhail Mishustin announced that the share of organizations planning to implement AI technologies has reached 77.2%. STC Module components integrated with Fplus servers are the first ready-made solution for a number of industries. We see a great need for such integrated products for image, object, and incident recognition, digital signal and image processing for CII, where Western-made equipment should not be used, as well as for working with large volumes of confidential or personal data, where speed is not important, but a secure environment is.
STC Module produces two types of cards: NM Quad 20 GB (large module) and NM Card Mini 5 GB (small module). SMT assembly of the cards is carried out in Russia, and they operate under the Astra Linux OS.
NM Quad and NM Card Mini GPU cards differ in that they do not have subsystems and blocks for rendering graphics. However, like GPU cards, they have specialized vector coprocessors for parallel operation.
Russian cards are currently slower than products from foreign vendors. However, the products from STC Module are significantly cheaper and more energy-efficient. Servers with them allow you to run neural networks that are trained on Western counterparts, but will not yet be able to handle AI model training.
STC Module is ready to produce up to 35 thousand cards per year at full capacity. They note that the products will be in demand in image, video, and biometrics processing systems, in solving problems of optimizing production processes, as well as for security, flaw detection, and diagnostic medicine systems.
There are many other scenarios for using the joint solution from Fplus and STC Module. For example, in the field of sales, artificial intelligence analyzes how sales have grown or fallen depending on various factors, and allows you to calculate the volume of future deliveries. In the banking sector, it will be possible to predict how a client will repay a loan, based on a scoring system that takes into account the borrower's current and past debts, the availability of property, place of work, and salary level. And Yandex Music analyzes the preferences and requests of the listener and other users with similar tastes and selects songs that are likely to appeal to a particular person.
Read related materials:
Russian Fplus has created the first crypto lock for secure server management
Aquarius has started designing servers for the Russian Amur microcontroller
