It is reported that due to a lack of the latest AI chips, coupled with insufficient power supply, generative AI research at South Korean universities is unable to operate normally.
Despite government support projects at Seoul National University and the Korea Advanced Institute of Science and Technology, they have received relatively small budgets, leaving them without sufficient funds to purchase Nvidia chips.
According to participating professors, building a model like OpenAI’s Sora requires hundreds of GPUs, which are difficult to obtain, forcing researchers to rely on older gaming GPUs.
Nvidia’s H100 chip costs around $40,000, while the A100 costs about $10,000. Using a system with eight Nvidia A100 GPUs, creating a Sora-like model would take 148 years.
Additionally, the power issue is crucial. Operating multiple GPUs requires a lot of electricity, which the universities cannot adequately supply.
Even if government budgets are approved, equipment purchases require approval and typically take more than three months. Research institutions reveal that general AI research needs at least eight GPUs, and servers cost more than $400,000 to $500,000.
Currently, in South Korea, Naver’s data center has 2,000-3,000 AI chips, followed by Samsung Electronics. However, this capacity is less than one-tenth of that of large American companies.
More importantly, even with money, it is challenging to buy AI chips, as companies that buy in bulk have priority, leaving smaller buyers out.
American universities obtain large GPU supplies through investments or donations. For example, the University of Florida built an AI supercomputer in 2020 with 1,120 A100 chips. The University of Florida is the alma mater of Nvidia co-founder Chris Malachowsky.
The University of Texas at Austin also has a large AI center, equipped with 600 H100 chips.
Disclaimer: This article is created by the original author. The content of the article represents their personal opinions. Our reposting is for sharing and discussion purposes only and does not imply our endorsement or agreement. If you have any objections, please contact us through the provided channels.