Carding 4 Carders
Professional
- Messages
- 2,730
- Reaction score
- 1,467
- Points
- 113
Find out how Russia is changing the game in the world of supercomputers.
Russia has announced a massive plan to build up to ten supercomputers by 2030. Each of them will potentially be equipped with 10,000 to 15,000 Nvidia H100 GPUs. From the point of view of computing capabilities, this number of processors will provide the country with performance at a level comparable to that used for training the Chat GPT system.
Such a system with such a large number of H100 GPUs could deliver 450 FP64 TFLOPS of performance, which is half the ExaFLOP. To date, this level of supercomputer performance has been achieved only in the United States.
The project is implemented by the Trusted Infrastructure team and is aimed at expanding the boundaries of computing capabilities. However, the required AI and HPC computing processors must be obtained from Nvidia, an American company.
The financial issue is also a problem for the Russian project. At current equipment prices, the project budget is about $ 6 billion. However, the rapid development of technologies in the future can reduce the cost of the project, and, possibly, by 2030, the cost of these systems can reach from 500 to 700 million dollars.
Currently, the most powerful supercomputer in Russia is Chervonenkis, owned by Yandex and equipped with 1,592 nodes with Nvidia A100 GPUs. It ranks 27th in the world in terms of computing power with a score of 21.53 PetaFLOPS. Three of the seven Russian supercomputers belong to Yandex, two to Sberbank, and one each to MSU and MTS. All of them run on previous-generation Nvidia GPUs.
Currently, there are only seven supercomputers in Russia that are among the world's top 500. For comparison, the US has 150 cars from this list, China - 134, Germany - 36, and Japan-33. As of June 2023, Russia ranks 12th in the rating.
Russia has announced a massive plan to build up to ten supercomputers by 2030. Each of them will potentially be equipped with 10,000 to 15,000 Nvidia H100 GPUs. From the point of view of computing capabilities, this number of processors will provide the country with performance at a level comparable to that used for training the Chat GPT system.
Such a system with such a large number of H100 GPUs could deliver 450 FP64 TFLOPS of performance, which is half the ExaFLOP. To date, this level of supercomputer performance has been achieved only in the United States.
The project is implemented by the Trusted Infrastructure team and is aimed at expanding the boundaries of computing capabilities. However, the required AI and HPC computing processors must be obtained from Nvidia, an American company.
The financial issue is also a problem for the Russian project. At current equipment prices, the project budget is about $ 6 billion. However, the rapid development of technologies in the future can reduce the cost of the project, and, possibly, by 2030, the cost of these systems can reach from 500 to 700 million dollars.
Currently, the most powerful supercomputer in Russia is Chervonenkis, owned by Yandex and equipped with 1,592 nodes with Nvidia A100 GPUs. It ranks 27th in the world in terms of computing power with a score of 21.53 PetaFLOPS. Three of the seven Russian supercomputers belong to Yandex, two to Sberbank, and one each to MSU and MTS. All of them run on previous-generation Nvidia GPUs.
Currently, there are only seven supercomputers in Russia that are among the world's top 500. For comparison, the US has 150 cars from this list, China - 134, Germany - 36, and Japan-33. As of June 2023, Russia ranks 12th in the rating.
