Intel, Naver form AI chip coalition to challenge Nvidia

Tech

Published: 11 Apr. 2024, 17:45

Intel, Naver form AI chip coalition to challenge Nvidia

Intel CEO Pet Gelsinger, left, and Ha Jung-woo, Naver's head of AI Innovation, speak on stage at the Intel Vision 2024 event in Phoenix, Arizona on April 9. [INTEL]

Intel and Naver formed an AI chip alliance, joining an array of global partnerships that aim to put a crack in Nvidia's dominance.

The two will set up a joint research lab and provide IT infrastructure for startups and academic institutions to conduct AI-related studies and development.

The partnership was announced at Intel Vision 2024 on April 9 in Phoenix, Arizona where Intel CEO Pat Gelsinger discussed the future of AI for enterprise.

Gelsinger introduced Naver as “one of the fastest growing Asian cloud service providers in the industry,” while Naver CEO Choi Soo-yeon, through a video message, showed confidence in creating a “new AI chip environment” with Intel — the world's third company to produce its own hyperscale generative AI model — at the event.

Naver will primarily focus on the Gaudi 2 chip where it will test and verify various AI software and build infrastructure based on it. It plans to create an open source platform which any developer, including startups and universities, can easily access to create their own AI services.

The joint lab, dubbed NICL (Naver Cloud-Intel-Co-Lab), will be joined by 20 research institutions including KAIST and Seoul National University.

The rising cost of Nvidia's AI chips and long lead time on orders has been causing trouble for the fast-growing global AI industry, especially for startups and universities with limited resources.

In addition, Nvidia's Compute Unified Device Architecture (CUDA) platform that works only with the American chip designers' silicon have locked in many developers into its expensive ecosystem.

Global IT companies are forming the alliance to counter that dominance.

Intel, Qualcomm, Google and many startups are reportedly joining forces to create an alternative to CUDA. OpenAI is also raising billion dollars to create its own ecosystem of AI chips in an attempt to lessen its reliance on Nvidia.

Samsung Electronics also announced recently that it is developing its own AI accelerator, called Mach-1, in collaboration with Naver.

The Korean firm said Thursday that its partnership with Intel won't affect its ties to Samsung.

Intel revealed more details of its Gaudi 3 AI accelerator at Vision 2024. First introduced last year, the Gaudi 3 is said to deliver 50 percent better inference and 40 percent better efficiency than the company's flagship H100 GPU at a lower cost.

The Gaudi 3 is expected to enter mass production in the second half of this year with TSMC's 5-nanometer nodes. It will then be supplied to Dell Technologies, HP and Lenovo, according to the company.

BY JIN EUN-SOO [jin.eunsoo@joongang.co.kr]