According to reports from foreign media on February 23rd, Vyacheslav Volodin, Chairman of the Russian State Duma, announced that the Saratov Oblast will construct Russia's largest supercomputer based on the data processing center being built by the Russian Savings Bank.
This is no secret, as Russia had previously stated plans to build 10 supercomputers by 2030, each equipped with 10,000 to 15,000 NVIDIA H100 GPUs.
The aim of this plan is to provide computational performance comparable to large language models like ChatGPT. However, the plan faces significant challenges, including overcoming technological limitations, scarcity of GPU supply, and the cost of computational power.
Currently, Russia's position in the field of supercomputing is relatively low, with its most powerful supercomputer, Chervonenkis owned by Yandex, ranking 27th globally in terms of computing power.
In contrast, the United States has 150, China has 134, Germany has 36, and Japan has 33 supercomputers. This makes Russia's goal of building 10 supercomputers appear ambitious, but also more challenging.
Despite the West's crackdown on Russia, the reality is that in the first 9 months of 2023, Russia purchased over $1.7 billion worth of chips from Western companies.
The top 20 companies sold $1.2 billion, including: AMD (including Xilinx), ADI, Intel (including Altera), Infineon, Macom, Marvell, Microchip, NXP, STMicroelectronics, Realtek, Texas Instruments, and others.
As for obtaining NV's AI chips, it's not difficult for Russia, though the official channels may not be as easy, third-party sources are readily available, albeit at a higher price. Of course, these matters are tacitly understood by these manufacturers.
