<Translated with Google Translate.>

Gwangju AI Data Center raises anchor to build domestic NPU farm

- KT‧Naver‧NHN cloud consortium project
- Gwangju CCTV integrated control center demonstration expected to expand nationwide
- Promoting public performance verification and AI semiconductor market entry​

In the AI data center, which is the core facility of the artificial intelligence-centered industrial convergence cluster (hereinafter referred to as the AI cluster) in Gwangju High-Tech District 3, the project to build an “AI semiconductor test and verification environment” that tests, verifies, and demonstrates domestic AI semiconductors is in full swing.

It is a demonstration project in which KT, Naver, and NHN Cloud build a neural network processing unit (NPU) using domestic AI semiconductors in an AI data center and provide cloud-based AI services, and synergies with the AI data center are expected.

In addition, by demonstrating domestic AI semiconductors at the CCTV integrated control center in Gwangju Metropolitan City, a basis for expanding application to CCTV integrated control centers nationwide is also prepared.​

According to the Artificial Intelligence Industry Convergence Project Group (Director Kim Joon-ha) on the 21st, a briefing session was held on the 21st at the auditorium of the Buk-gu AI Project Center in Gwangju as part of the 'K-Cloud Project' hosted by the Ministry of Science and ICT. announced the start of the support project.

KT Cloud, Naver Cloud, and NHN Cloud, which are cloud companies, participate in the project to create an AI semiconductor test and verification environment, and Rebellion, Sapion Korea, and Furiosa AI participate as AI semiconductor companies.

This project was promoted by securing a budget of 20 billion won for the project of “creating an AI semiconductor test and verification environment” for two years from 2023 to 2024.

The ‘AI Semiconductor Testing and Verification Environment Creation’ project proceeds with support tasks such as ▲establishing a domestic AI semiconductor server farm ▲proving AI services in the public sector ▲development of an operating platform.

Currently, most domestic AI semiconductors are in the pre-commercialization stage, and in order to enter the market, it is essential to verify not only comprehensive performance but also compatibility with existing systems and stability.

Through this project, it plans to build an NPU farm with an amount of about 20 petaflops (PF) of computation (1PF performs 1,000 trillion floating-point operations per second) within the AI data center.

'NPU' is an AI semiconductor that has improved large-scale power consumption, a major drawback of the GPU (graphics processing unit) method used by Nvidia of the United States, which is preoccupying the world's mass computing AI semiconductor technology.

In addition, an empirical verification system will be established to prepare a one-stop verification system that can verify the compatibility, stability, usability, scalability, and performance of AI semiconductors.

'Establishment of AI semiconductor verification system' supports early market entry by easing the investment burden of AI semiconductor manufacturers by establishing a one-stop test environment for compatibility, stability, usability, scalability, and performance of domestic AI semiconductors in the public domain. and to strengthen the competitiveness of the domestic AI semiconductor market.

Through this, the AI Industry Convergence Project Group expects that it will be of great help to domestic AI semiconductor companies that have AI semiconductor technology but are struggling because they have not been able to secure AI service application performance.

It is expected to contribute to minimizing trial and error in the actual service stage by providing references (cases) that enable entry into demand markets through functional and performance verification of domestic AI semiconductors before commercialization such as cloud services.

Director Kim Joon-ha said, “If a cloud environment based on domestic AI semiconductors is established within the AI data center, it will help advance into the early market of AI semiconductors, which require high costs from the pre-commercialization stage.” Through the demonstration of the center, we expect to lay the groundwork for applying domestic AI semiconductors to CCTV integrated control centers nationwide and to promote competitiveness in the domestic AI semiconductor market.”