Intel Unveils Next-Generation AI Chip 'Gaudi3'... Challenges Nvidia

U.S. semiconductor company Intel unveiled its self-developed artificial intelligence (AI) chip 'Gaudi3,' challenging Nvidia.

According to foreign media on the 11th, Intel revealed its latest AI chip 'Gaudi3' at the 'Intel Vision' event held on the 9th (local time) in Phoenix, Arizona, USA. This came just four months after showcasing the Gaudi3 prototype at a new product launch event held in New York last December.

Intel's next-generation AI chip 'Gaudi3'.
[Photo by Intel]

원본보기 아이콘

Intel explained that Gaudi3 has more than twice the power efficiency of Nvidia's latest chip, the H100 graphics processing unit (GPU), and can run AI models 1.5 times faster. It also claimed that in large language model (LLM) training, Gaudi3 is up to 1.7 times faster than the H100 and up to 1.3 times faster than Nvidia's H200 used for inference. Intel reported testing Gaudi3 with Meta's open-source LLM 'Llama' and the open-source LLM 'Falcon' built by an AI company in Abu Dhabi, United Arab Emirates (UAE).

Intel announced that Gaudi3 is scheduled to be released in the third quarter, and U.S. server companies Dell, HP, and Supermicro will build Gaudi3 systems.

Intel did not disclose the price of Gaudi3 but described it as "competitive." Intel's Vice President of Software, Das Kamhout, stated, "We expect (Gaudi3) to be very competitive compared to Nvidia's latest chips." He added, "Competitive pricing, differentiated open integrated network on chip, and the use of Ethernet, an industry-standard network technology," as reasons why they believe Gaudi3 is a powerful product.

With Intel releasing a new chip comparable in performance to the H100, competition surrounding AI chips is expected to intensify. Nvidia, which currently holds 80% of the global AI chip market, unveiled the B100 and B200 GPUs based on the new 'Blackwell' platform last month as successors to the H100. These products are scheduled for release in the second half of this year.

U.S. semiconductor company AMD launched the next-generation data center GPU MI300X last year and has already secured major customers such as Meta and Microsoft (MS).

Intel is forming an 'anti-Nvidia front' with Qualcomm, Google, and others, also working on building open software for AI app development. This is to challenge Nvidia's CUDA, an AI-related app development support software platform.

© The Asia Business Daily(www.asiae.co.kr). All rights reserved.