[Exclusive] 'Precious' GPU, Poorly Managed at Research Institutes

NST Audit Points Out Management Issues at Life Science, Chemistry, and Food Research Institutes
Assets Purchased for Millions Not Managed as Assets but Treated as Consumables
Food Research Institute Staff Operates Coin Mining Server Using Institute-Owned GPU

Major government-funded research institutes in the domestic science and technology sector were found to have experienced illegal usage of expensive NVIDIA graphics processing units (GPUs) due to improper management despite purchasing them. In an era where GPUs are essential for research in artificial intelligence, it was confirmed that some researchers failed to properly utilize even the GPUs they had purchased. As the use of GPUs in scientific research is increasing, there appears to be a need to establish a more transparent management system.

A cryptocurrency mining server illegally installed within the institute by an employee of the Korea Food Research Institute using the institute's asset, a GPU. Photo by NST

According to the Audit Committee of the National Science and Technology Research Council (NST) on the 5th, recent issues related to GPUs were pointed out at the Korea Research Institute of Bioscience and Biotechnology, Korea Research Institute of Chemical Technology, and Korea Food Research Institute.

According to NST, the Korea Research Institute of Bioscience and Biotechnology purchased GPUs worth 125.22 million KRW from 2021 to 2023. The institute explained that "GPU purchases increased due to operating the National BioResearch Resource Information Center."

However, the Bioscience Institute did not register the purchased GPUs as assets but treated them as computer consumables, excluding them from management targets. As a result, it was impossible to track the usage status and location of GPUs in the ledgers.

The Korea Research Institute of Chemical Technology also purchased GPUs worth 826.14 million KRW during the same period but was criticized for inappropriate asset management as they were also treated as consumables. The Chemical Institute explained that many GPUs were purchased to operate the Chemical Data-Based Research Center.

Recently, each research institute has been placing more emphasis on GPUs than central processing units (CPUs) when securing computing equipment. GPUs are being used in digital twins, virtual simulations, and more. In the case of the Chemical Institute, the amount spent on GPUs was about 2.5 times higher than that on CPUs. This phenomenon arose as GPU purchases increased and prices rose.

NST found through on-site investigations that there were no issues with the utilization of GPUs purchased by the Bioscience and Chemical Institutes, but pointed out that detailed management was not conducted on paper. NST advised establishing management and supervision measures so that it can be confirmed on which equipment the GPUs are installed and used, and when and how they were disposed of. Both institutes responded that they would ensure thorough documentation management even for purchases of GPUs and CPUs exceeding 1 million KRW.

There is also a view that such problems cannot be seen as issues of specific research institutes alone. Since many institutes purchase expensive GPUs, management is also important. The case of the Korea Food Research Institute is representative.

According to the NST audit, a director at the Food Research Institute was found to have personally set up a cryptocurrency mining server within the institute using 12 expensive GPUs owned by the institute. Despite installing air conditioning and an access control system with the institute’s budget to manage the mining server, this fact went unnoticed for several years. NST estimated that this illegal activity caused losses of about 78 million KRW, including electricity bills and equipment depreciation costs. If accurate grounds for GPU purchases and ledger management had been in place from the beginning, such incidents would not have occurred.

An NST official explained, "While investigating the actual situation targeting institutes with many purchase records, some problems were revealed," adding, "Since GPUs are expensive equipment, thorough management is necessary."