Nvidia challenger Sapeon unveils new AI chip for data centers

Sapeon Inc., a semiconductor startup backed by South Korea’s second-largest conglomerate SK Group, on Thursday unveiled a new artificial intelligence chip for data centers to join the global race in the sector and challenge the world’s chip design leader Nvidia Corp.

Sapeon introduced the X330 neural processing unit (NPU) for inference, which quadruples the computing performance of its existing model X220 and doubles the power efficiency.

Compared with competitors’ inference chips released this year, the X330 provides double the computing performance and its power efficiency is more than 1.3 times better, the startup headquartered in Santa Clara, California, said.

An internal analysis showed the X330’s power efficiency was better than graphics processing units with the same specifications by 1.3 to 1.9 times, said Sapeon Chief Technology Officer Michael Shebanow.

Sapeon, spun off from South Korea’s top mobile carrier SK Telecom Co. in April 2022, designs AI semiconductors for data centers. The market is currently dominated by Nvidia of the US.

MASS PRODUCTION IN H1 2024 THROUGH TSMC’S 7 NM PROCESS

Sapeon plans to mass-produce the X330 in the first half of 2024 through the 7-nanometer process of the global foundry leader Taiwan Semiconductor Manufacturing Company (TSMC) after testing its prototype with customers and verifying its reliability.

The X330 supporting large language models is capable of running an LLM based on a transformer, a source technology of conversational AI ChatGPT developed by Microsoft Corp.-backed OpenAI based on generative AI technology.

The latest AI chip for inference, which is set to implement AI services based on learning data, has a wider application range than its previous models, making it easy to be used in various industries, Sapeon said.

The startup plans to provide open neural network exchange-based software stacks along with the X330 when it is installed on servers to optimize performance while offering AI inference platforms and software development kits.

“We started a new journey today with the X330, the best AI processor for inference,” said Sapeon Chief Executive Officer Ryu Soojung at the event to unveil the new chip. “We will develop Sapeon into a company that shares a convenient life with AI, as well as the era of ‘AI Everywhere’ and ‘AI Big Bang.'”

In addition to the X330, Sapeon aims to introduce chip designs for self-driving automobiles and AI NPUs for high-performance edge devices, which provide an entry point into enterprise or service provider core networks.

Sapeon also plans to expand its cooperation with SK Telecom to actively target new service markets while protecting customers from security threats with systems using AI chips to defend against voice phishing, smishing and spamming.

Write to Seung-Woo Lee at leeswoo@hankyung.com
 
Jongwoo Cheon edited this article.