At the launch event for XHEART's New Chip Journey, Momenta CEO Cao Xudong publicly unveiled, for the first time, the company's third significant insight: large models shape chip design. Previously, Momenta had already introduced two insights focused on achieving large-scale L4 autonomous driving: solving long-tail challenges through data-driven methods and acquiring vast amounts of data through large-scale production.

In late 2022, Momenta introduced the concept of large models shaping chip design and partnered with XHEART to bring this vision to life. Cao Xudong emphasized that chips designed specifically for large models can deliver a tenfold performance boost compared to general-purpose chips. Achieving large-scale L4 autonomy necessitates models with tens to hundreds of billions of parameters. However, general-purpose chips often struggle to simultaneously satisfy the three critical constraints of power consumption, automotive-grade reliability, and cost—a challenge known as the "impossible trinity" in chip design.

Cao Xudong further explained that only by tailoring chip design to the specific requirements of large models can we support models with hundreds of billions of parameters within these constraints, paving the way for large-scale L4 autonomy and driving breakthroughs in robotics and other physical AI applications.

During the event, XHEART New Chip Journey CEO Li Zonglin officially introduced the new X7 Ultra chip, specifically engineered for large models. This chip features an ultra-large core NPU architecture, significantly enhancing data interaction and interconnection rates within the core to terabits per second (TB/s) levels, far surpassing inter-core data interactions. Additionally, it incorporates China's first mass-produced Armv9 automotive-grade CPU, offering a 20% performance increase and a 10% improvement in energy efficiency.