Russia has unveiled plans to create up to ten supercomputers by 2030, a development that could significantly enhance the country’s computational capabilities. Each supercomputer is expected to be equipped with 10,000 to 15,000 NVIDIA H100 graphic processors, providing a level of performance comparable to the system used to train the Chat GPT artificial intelligence model.
With a large number of G100 graphic processors, each system has the potential to deliver a performance of 450 FP64 TFLOPS, which is half the Exaflop. Currently, only the United States has achieved this level of supercomputer performance.
The Trusted Infrastructure command is leading the project with the aim of expanding computational boundaries. However, the project faces challenges in terms of sourcing the required computing processors for artificial intelligence and high-performance computing from NVIDIA, an American company.
Finances also present a hurdle for the Russian project, with the estimated budget currently standing at around $6 billion. However, as technology rapidly develops, it is anticipated that the cost of the project could decrease. By 2030, the cost of these systems is projected to range from $500 to $700 million.
The most powerful supercomputer in Russia currently is Chervonenkis, owned by Yandex. Equipped with 1,592 nodes containing NVIDIA A100 graphic processors, it ranks 27th in the world in terms of computing power with an output of 21.53 Petaflops. Additionally, Yandex owns three of the seven Russian supercomputers, while Sberbank owns two, and Moscow State University and MTS each own one. All of these supercomputers utilize previous generations of NVIDIA graphic processors.
Russia currently has only seven supercomputers listed in the world’s top 500, with the country ranking 12th globally as of June 2023. By comparison, the United States has 150 supercomputers listed, China has 134, Germany has 36, and Japan has 33.