[Reading Science] The Era of Global AI Leadership Competition... 'Supercomputers' Become Key Weapons

UK Prime Minister Directly Explains Significance of Supercomputer Adoption
Tesla Develops Supercomputer 'Dojo'
German-French Alliance Introduces Exascale Supercomputer
KISTI's 6th Supercomputer Adoption Delayed Due to GPU Price Surge

UK Prime Minister Rishi Sunak directly mentioned the introduction of supercomputers on the 2nd (local time). Through his X (Twitter) account, Prime Minister Sunak spoke about the UK Department for Science, Innovation and Technology's plan to introduce a new supercomputer.

"AI models are becoming increasingly powerful. This investment will equip UK science with the tools needed to create the most advanced AI safety models. Supercomputers will strengthen the UK's position in the field of AI safety."

It is uncommon for the UK Prime Minister to directly address the development of supercomputers, which is neither a political nor economic issue. Sunak emphasized the large-scale introduction of supercomputers in conjunction with the ‘1st AI Safety Summit’ that he personally conceived. He declared that supercomputers are essential to taking the lead in the AI era. Sunak proposed international discussions on AI stability and reliability while simultaneously laying the groundwork for large-scale domestic investment.

Through this investment, the UK plans to install new supercomputers in Cambridge and Bristol respectively. The new supercomputers are said to have more than 30 times the capacity of the current largest public AI computing capacity in the UK. When the supercomputers become operational next year, the UK will gain a powerful ally for AI development.

Elon Musk’s Tesla recently attracted attention not with electric vehicles but with a supercomputer named ‘Dojo.’ Investment bank Morgan Stanley predicted that Dojo, used by Tesla for autonomous driving learning, could increase Tesla’s market value by about $500 billion (approximately 665.5 trillion KRW), drawing market attention. Tesla is developing supercomputers to process large volumes of autonomous driving data.

Supercomputers reflect the national power and corporate capabilities of countries worldwide. The UK plans to use the newly acquired supercomputers not only for AI but also to lead innovation in new drug development and clean energy sectors. Although the UK ranks sixth in the world supercomputer ownership rankings announced in June, following the US, China, Germany, Japan, and France, it aims for a higher position.

▲Perspective view of Supercomputer No. 5.
[Photo by KISTI]

Traditionally, the US has led the supercomputer field. The US develops the core CPUs and GPUs used in supercomputers and leads in clustering technology that connects these components. China has risen beyond an emerging power to a level that pressures the US. However, recent semiconductor export restrictions to China have cast a red light on the growth of China’s supercomputing capabilities.

◇ Soaring GPU prices disrupt the introduction of South Korea’s 6th supercomputer = According to the TOP500 list ranking supercomputers, the November ranking of the world’s supercomputer performance was released, with the Oak Ridge National Laboratory’s Frontier maintaining its number one position since June. Frontier is an exascale supercomputer.

In South Korea, eight supercomputers are recognized, including Samsung Advanced Institute of Technology’s ‘SSC-21’ (20th) and ‘SSC-21 Scalable Module’ (387th), the Korea Meteorological Administration’s ‘Guru’ (37th) and ‘Maru’ (38th), SK Telecom’s ‘Titan’ (47th), Korea Institute of Science and Technology Information (KISTI)’s ‘Nurion’ (49th), KT’s ‘KT DigiX SuperPOD’ (58th), and Gwangju Institute of Science and Technology (GIST)’s ‘Dream-AI’ (207th).

Supercomputers must be replaced when their lifespan ends, requiring large-scale funding each time. Supercomputers are built by clustering general-purpose CPUs from companies like Intel and AMD on a large scale. However, the situation has changed as graphics processing units (GPUs) have emerged as key components.

Since the emergence of generative AI ChatGPT, Nvidia GPUs have become difficult to obtain, and their prices have soared to unprecedented levels. Nvidia’s latest GPUs cost over $30,000 each. Hundreds to thousands of these GPUs are needed to build a proper supercomputer.

The HP supercomputer ‘Isambard AI’ that the UK plans to install will use 5,000 of Nvidia’s advanced AI chips. To accommodate this, the UK had to increase its supercomputer procurement budget by 200% from the original ?100 million.

Germany and France have joined forces. The two countries agreed to jointly cooperate on the ‘Jupiter’ project to introduce the European Union’s first exascale supercomputer. The Jupiter computer alone will cost a total of €273 million (approximately 386.6 billion KRW). Jupiter will be the first European system to exceed one billion calculations per second and will be used for computations in areas such as climate change, infectious diseases, and nuclear fusion energy. It is also expected to play an active role in AI.

The immediate impact of soaring GPU prices has also hit South Korea. The introduction of KISTI’s 6th supercomputer is in crisis. Following the 5th supercomputer ‘Nurion,’ the 6th passed a preliminary feasibility study in 2021 with a budget of 299.9 billion KRW allocated, and installation was scheduled for next year.

The 6th supercomputer aimed to have a computing power of 600 PF (petaflops, 600 quadrillion operations per second), about 20 times faster than ‘Nurion,’ which had 25.7 PF performance. However, no companies have responded to multiple bids. Companies are not bidding for the 6th supercomputer because the budget cannot cover the soaring GPU prices. The rising won-dollar exchange rate has further increased the required budget. As a result, KISTI lowered the required performance in the third bidding announcement.

KISTI expects that if the first three bidding announcements fail, a fourth round allowing single bids might attract challengers, but it is uncertain whether any company will take on the challenge amid GPU shortages and price increases. Even if the price is met, timely delivery is not guaranteed. High-end GPUs cannot be purchased immediately even if money is offered.

Kim Jae-su, director of KISTI, said at the 35th anniversary event of supercomputer introduction last August, "The introduction of the 6th supercomputer is essential, but we are facing the worst environment with soaring GPU prices, interest rate hikes, and high exchange rates."

The budget is already insufficient, and the purchase timing has been delayed. Lee Jong-ho, Minister of Science and ICT, expressed willingness to increase the budget, but it is uncertain whether the introduction will take place next year. The academic community is already concerned about declining research morale due to delays in supercomputer introduction.