Russia Wants to Build Multiple Top 10 Supercomputers by 2030

Russia was an early leader in high-performance computing, but it has failed to keep up with supercomputer programs in other nations. While the country is still embroiled in a war with Ukraine, the Russian government's Trusted Infrastructure team has announced plans to build a cadre of new supercomputers by 2030 that could push Russia to the forefront of the booming artificial intelligence industry. The problem, however, is the war has cut Russia off from the powerful AI accelerators it will need for the project.

Today, Russia has just seven supercomputers on the Top 500 list, a far cry from the 150 residing in the US and China's 134. Russia's most powerful current supercomputer is the Chervonenkis system from tech firm Yandex. Chervonenkis (above) runs at 23.53 petaflops with 1,592 Nvidia A100 GPUs. Other high-performance systems like Lyapunov (also at Yandex) and Moscow State University's Lomonosov run on older-generation Nvidia cards.

The Russian plan is not just to get more computers—the team aims to crack the top 10 with speeds ranging from 400 to 500 FP64 teraflops, according to Tom's Hardware. That's about half as fast as the exaflop machines leading the world, like the US Department of Energy's Frontier computer. Germany is planning an ARM and Nvidia-based exascale supercomputer, as well. To make this happen, the government would need around 10,000-15,000 AI accelerator modules from Nvidia's latest H100 line. That's per system, not overall. That's no short order for a country hit with economic sanctions after the Ukraine invasion. Nvidia is a US company subject to export controls enacted by the US government.

Russia may find it difficult to get enough AI accelerators in the current geopolitical climate. Credit: Nvidia

If the Russian government can get its hands on that many cards, it would have a tremendous amount of processing power at its disposal. Each proposed computer would be about as fast as the system OpenAI used to train its ChatGPT algorithm. OpenAI is also based in the US, so its products can be restricted in Russia.

Experts who have examined Russian military technology say the country is still finding ways to get western chips it cannot replicate domestically, reports CNBC. These gray market parts are most likely coming into Russia via intermediaries like Turkey and Syria. However, securing tens of thousands of GPUs that are already in short supply is a much taller order. In addition, buying the cards through a third party will inflate the price, and the MSRP on tens of thousands of H100 GPUs is already $6 to $7 billion. As Russia continues to wage a costly war in Europe, this project could easily fall by the wayside.