Russian authorities hope to step up the development and manufacture of supercomputers. However, high costs and the sanctions slapped on Russia over its invasion of Ukraine make the supercomputer program a challenging enterprise.

The idea of supercomputers seems to be high on the Russian leadership's agenda. Earlier this month, president Vladimir Putin instructed the government to develop and implement measures to increase the computing power of supercomputers in Russia, the Kremlin website reported.

"To develop and implement a set of measures aimed at increasing the computing power of supercomputers located in the Russian Federation, defining specific parameters for increasing these capacities," the report on the Kremlin website reads, adding that by March 1, the government has to develop tools for using the archives of state and municipal bodies and library funds to create data sets. Those data are expected to be subsequently fed to supercomputers.

The main idea behind supercomputers is that they could facilitate much faster calculations than normal machines. Unlike conventional computers, supercomputers perform tasks simultaneously rather than sequentially. Supercomputers can be used in various fields, from the creation of new medicines to the development of new technologies.

 

Plans for the supercomputer race

The idea of strengthening performance in the area of supercomputers has been floated in Russia for a few months.

In late 2023, the quasi-government organization Digital Economy said that Russia could build 10 new supercomputers with a capacity of 10,000-15,000 graphics processing units (GPUs) by 2030. Digital Economy used H100, Nvidia's latest generation GPUs, released in 2022, as a reference for chips that would power future Russian supercomputers. Each H100 chip contains 80bn transistors, providing 56-66 teraflops of performance.

A spokesman for Digital Economy told the Russian business newspaper Vedomosti that the organisation is in close contact with the business community, collecting opinions and proposals on the topic of data economy in a bid to facilitate the supercomputer programme.

Previously, Russia has already built supercomputers. Currently, seven Russian supercomputers are included in the world's Top 500 rating. But this number pales in comparison with that of the rating's leader, the USA, which has 150 supercomputers in it, or runner-up China, which has 134. Germany, which is in third place, contributed 36 supercomputers to the rating and is followed by Japan with 33. In the June 2023 rating, which is the most recent one available, Russia is in 12th place in the world by the number of supercomputers.

However, most of these Russian supercomputers have been developed and operated by private companies, with little government involvement in the process. Of the Russian supercomputers, three (Lyapunov, Chervonenkis and Galushkin) belong to the leading Internet company Yandex, two (Christofari and Christofari Neo) to the biggest lender Sberbank and one to mobile phone operator Mobile TeleSystems (GROM). Only one supercomputer, Lomonosov, was built and is operated exclusively by a government institution, Moscow State University.

 

Ambitious plans

Today, a cluster with 15,000 Nvidia H100 graphics cards would be one of the most productive in the world, Sergei Pimkov, Deputy General Director of Product Development and Operations at the tech firm Selectel was quoted as saying by Vedomosti.

According to Pimkov, such computing power is required for artificial intelligence (AI) models, such as GPT4. "Over time, the stakes only grow, so such supercomputers are definitely needed for AI training," he said.

Yandex uses machine learning methods to better help people to solve their tasks, a company spokesman told Vedomosti. "Thanks to them, we get fast and accurate answers in Search, talk to [voice assistant] Alice like a live person, watch videos immediately in our native language and do many other things," he added.

One of the most common scenarios for using supercomputers is training large language models that allow for the most advanced AI-based chatbots, Pimkov said. "In general, supercomputers are needed when it is necessary to perform the most intensive calculations - for forecasting weather and climate conditions, engineering simulations, mathematical calculations, synthesis of new materials, space modelling," he added.

 

Sanctions as an obstacle

Still, there's one important factor that might put a lid on Russia's ambitious plans for supercomputers. All supercomputers currently operated in Russia were built before Russian invasion of Ukraine and ensuing Western sanctions and relied heavily on chips provided by the world's leading manufacturer Nvidia.

For instance, Yandex's most powerful supercomputer, Chervonenkis, has 1,592 nodes with Nvidia A100 GPUs, ranking 27th in the world in terms of computing power. However, in September 2022 the United States imposed restrictions on exports to Russia and China of a number of GPUs manufactured by AMD and Nvidia (including H100, the newer generation).

H100 GPUs are on average 1.66 times faster than A100 GPUs, and 10 supercomputers with a total of 15,000 GPUs powered by chips comparable to H100 would bring Russia to the top of the world supercomputer ranking. But where and how Russia is going to get chips comparable to H100 to power its new supercomputers, remains anyone's guess.

 

Who will foot the bill?

In addition, even if Russia were able to ensure the supply of H100-type chips in quantities sufficient for its ambitious plans for supercomputers, the issue of funding is likely to become rather acute.

If Russia wants to build higher-capacity supercomputers that could be competitive against their analogues built in other countries, it has to come up with a pretty hefty sum.

At today's prices, 10 computers with 10,000-15,000 GPUs will cost $6bn, Maxim Koposov, general director of the Promobit tech firm, was quoted as saying by Vedomosti.

According to Koposov, it is likely to be very difficult to bring in such a quantity of chips unnoticed by the manufacturer, given the current ban. "Such sums would be more appropriate to invest in the creation of Russian GPUs, from development to production and at least in partnership with friendly countries," he added.

 

New national program

Meanwhile, Russia is preparing a new national project for technology-based economic development that could technically be used to fund the supercomputer programme.

In 2024, Russia will launch a national project aimed at developing the economy on the basis of data generated in digital systems - both business and government, the Kremlin has announced.

The exact size of funding that Russia's supercomputer programme could collect under the tech national project is hard to estimate at this point. But, according to Russian senior government officials, substantial sums are slated to be allocated to the development of a related field, quantum computing.

According to deputy prime minister Dmitry Chernyshenko, as much as RUB41bn ($461.6mn) will be allocated through 2026, with the cash coming from both public and private sources.

In the past, however, national projects launched by the Russian government, often proved to be inefficient, poorly managed and marred by corruption scandals. Currently, things are likely to be yet more complicated due to Russia's war in Ukraine and substantial military spending.