Intel takes fight to Nvidia with Core Ultra release, AI accelerator

Intel unveiled its first neural processing unit (NPU)-topped PC processor, the Intel Core Ultra, on Monday in Korea, throwing down the gauntlet in a market largely dominated by Nvidia.

Intel Korea CEO Kwon Myung-sook poses with the Intel 4 process-based wafer [INTEL KOREA]

Intel unveiled its first neural processing unit (NPU)-topped PC processor, the Intel Core Ultra, on Monday in Korea, throwing down the gauntlet in a market largely dominated by Nvidia.

The American chipmaker also unveiled the Intel Gaudi 3, an AI accelerator tailored for large-language models (LLMs) which is slated for launch next year.

"As demand for digital transition is soaring across all industries, the digital economy, which accounts for 15 percent of current GDP, will increase to 25 percent in 10 years," said Kwon Myung-sook, CEO of Intel Korea in a press conference Monday.

"Intel will support not only hardware but also software solutions for AI so that clients can establish AI infrastructure for clouds, network, PC and edge devices."

The Intel Core Ultra is a rebranded version of the company's widely known i series which started in 2008 with the Core i7-900. The Core Ultra has undergone a major revamp in its architecture for the first time in 40 years, the chipmaker said.

The seven-nanometer chip has been newly topped with an NPU, the type of chipset that emulates the neural networks of a human brain. They are said to be more suitable for processing AI compared to graphic processing unit (GPU) because they are more purpose-built.

Intel said the NPU deployment helped boost the processor's energy efficiency by 2.5 times compared to the previous version.

The Core Ultra will be installed on Samsung Electronics' first AI laptop Galaxy Book4 slated for launch next month as well as LG Electronics' Gram 16.

The fifth generation Xeon server processor was also revealed, also topped with an NPU. Its basic inference capability has been increased by 21 percent and AI inference by 42 percent compared to the previous generation.

The Gaudi3, specifically designed for generative AI models, boasts a fourfold increase in data processing speed compared to its previous version. It will compete with Nvidia's H200 and AMD's MI300X.

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