On March 2, the Beijing AI Public Computing Platform announced another expansion, with its intelligent computing power exceeding 10,000P. This makes it Beijing's largest and a nationally leading ultra-large-scale, high-performance single-unit intelligent computing cluster. This will strongly support innovative entities in integrated training and inference for trillion-parameter general-purpose foundational models. It's reported that 10,000P of computing power is equivalent to the combined capabilities of five million high-performance notebooks.

A section of the Beijing AI Public Computing Power

[Photo by the Publicity Department of the CPC Haidian District Committee]

With breakthrough advancements in AI technology, enterprises' demand for computing power continues to surge. Effective computing power has become both the core driving force and the critical scarce resource in cutting-edge technology-driven innovation. The 10,000P Beijing AI Public Computing Platform has achieved multidimensional breakthroughs in efficiency, stability, and scalability through top-level unified architecture design. It not only significantly enhances the training efficiency of trillion-parameter models, but also establishes a full-link monitoring and self-healing system. This system can predict potential hardware failures in advance, ensuring an annual training task completion rate of over 99 percent. This will effectively help enterprises break free from computing power constraints, allowing them to focus on overcoming core technical challenges.

The Beijing AI Public Computing Platform started construction on September 15, 2023, and successfully activated a 3,500P-scale high-performance computing cluster in March 2024.

(Source: Beijing Daily)