Naver, Intel, KAIST "Proving the Possibility of an AI Ecosystem Based on Gaudi.. Need for Competition"
- Input
- 2025-05-29 15:56:55
- Updated
- 2025-05-29 15:56:55
"With the help of Intel and Naver, we have published or are reviewing a total of 40 papers from 22 university labs over the past year. We also developed a Gaudi software inference model that processes tokens up to 1.2 times more than the NVIDIA GPU A100" (Seong Min-hyeok, Professor of Computer Science at KAIST)
"Recently, national AI projects seem to be designed only around semiconductor suppliers, and service companies seem to be neglected. I hope that Intel will support a lot so that this cooperation can continue, benefiting academia and providing Naver Cloud with an opportunity to compete from a broader perspective" (Lee Dong-su, Executive Vice President of Naver Cloud AI Computing Solutions)
■ Building an AI Ecosystem Based on Gaudi 2.. Verified through Papers and Practice
Intel, Naver Cloud, and Korea Advanced Institute of Science and Technology (KAIST) held a press briefing on the 29th in Yeongdeungpo-gu, Seoul, to share the achievements of the AI Joint Research Center. Previously, Intel and Naver Cloud announced in April last year that they would build a software ecosystem based on 'Gaudi' through joint research with KAIST and other domestic academia and startups.
Subsequently, 22 labs from KAIST, POSTECH, and Seoul National University formed an AI Joint Research Center and were each provided with a Gaudi 2 node for AI research. As a result, they achieved △publication of 15 papers and submission of 25 papers △release of over 30 open-source projects on GitHub repositories △development of Gaudi vLLM equivalent to NVIDIA GPU A100 △performance verification of Gaudi 2 at Naver Cloud data center. However, this project is scheduled to end its first phase in August, and the continuation is yet to be determined.
Kim Jeong-ho, a professor of electrical and electronic engineering at KAIST, known as the 'Father of HBM', said, "Currently, the AI ecosystem is centered around NVIDIA, but through this new experience of breaking it once, we aim to grow Intel and the Korean AI semiconductor and AI industry together," adding, "Our lab is designing HBM with AI, and we are doing well using Gaudi 2, and the students are satisfied."
■ "Competition over NVIDIA Monopoly.. Service Companies Excluded"
During the briefing, concerns were also raised about the government's plan to configure the national AI computing center solely with NVIDIA GPUs such as H200 and B200.
Cho Min-seong, Director of Corporate Strategy Sales Division at Intel Korea, said, "If NVIDIA raises (GPU) prices in a state without competition, we have no choice but to match them," adding, "Supply issues may also arise, so competition is necessary for national competitiveness, even if it's not Intel."
Professor Kim also stated, "There are various elements such as ecosystem dependency and openness, but it seems that growth occurs through competition," adding, "I hope the cooperation model of Intel, Naver, and KAIST continues and becomes a successful case of various ecosystems."
Additionally, Executive Vice President Lee expressed regret that the government shows little interest in service companies like Naver when fostering AI businesses.
Executive Vice President Lee said, "It seems that our service companies do not have the opportunity to directly convey their requirements, as they are not even invited to national events and instead visit specific semiconductor companies," adding, "I think it will be extremely difficult to create such a good ecosystem while service companies continue to be excluded."
solidkjy@fnnews.com Koo Ja-yun Reporter