Momenta's third insight: large models redefine chip design

At today's XHEART New Chip Journey event, Momenta CEO Cao Xudong revealed the company's third key insight: large models are redefining chip design. This follows earlier insights focused on solving long-tail issues through data-driven approaches and achieving massive data through mass production. In late 2022, Momenta introduced the concept of 'large models defining chips' and began collaborating with XHEART. Cao stated that chips designed for large models offer a 10x performance boost over general-purpose chips. He emphasized that achieving scaled L4 autonomy requires models with hundreds of billions to trillions of parameters, which general-purpose chips struggle to support due to the 'impossible triangle' of power consumption, automotive-grade reliability, and cost. He argued that only by designing chips from the needs of large models can these constraints be overcome, enabling scaled L4 and advancing physical AI applications like robotics. Earlier today, XHEART CEO Li Zonglin launched the X7 Ultra, a chip specifically designed for large models. It features a massive NPU architecture with significantly faster intra-core data transfer and is the first Armv9 automotive-grade CPU in China, offering a 20% performance boost and 10% better energy efficiency.