In 2025, the Russian company Fplus will start producing servers with neural accelerators developed by the STC “Module”. Fplus plans to release about 100 such servers in 2025, and by 2026-2027 to increase production volumes to several thousand devices. Will the devices be able to replace Nvidia products? IT-World investigated.
Fplus servers with a GPU card are designed to work with neural networks, process signals and images. However, they cannot be used to train artificial intelligence.
The project uses two types of GPU cards: NM Quard with 20 GB of memory and NM Card with 5 GB. Card production will be localized in Russia, and the annual volume can reach 35,000 units. All solutions will work on the Russian operating system Astra Linux, Vedomosti learned.
Alexey Melnikov, Managing Partner of the Fplus IT holding, emphasizes the demand for solutions for processing large volumes of data and neural network tasks in the Russian market, especially for critical information infrastructure, which requires the use of exclusively domestic equipment.
The Fplus Voskhod server with NTC Modul cards is included in the register of domestic products of the Ministry of Industry and Trade.
At the moment, there is a shortage of servers, especially with the integration of solutions for artificial intelligence, which is due to Nvidia’s monopoly on the global GPU market. However, Russian companies are actively looking for alternatives and working on creating their own solutions compatible with AI accelerators, Fplus reported.
Thus, a server from the Graviton company, capable of supporting up to five graphics accelerators, appeared in the register of the Ministry of Industry and Trade. And the company is also working on several more platforms for AI, the appearance of which is expected next year.
NTC Modul, in turn, has been developing high-performance processors and systems for aviation and space, as well as solutions in the field of neural network technologies since the mid-1990s.
However, according to experts, servers with GPU cards from NTC Modul cannot yet completely replace Nvidia solutions, especially in neural network training tasks. Nevertheless, the company’s products will be in demand, especially in government and corporate structures. The solutions will be in demand in critical systems where the use of Russian components is determined by legal requirements.