On October 24, the world's first ultra-small, mobile brain-inspired intelligent computing device, "BI Explorer" (BIE-1), was officially launched in the Hengqin Guangdong-Macao Deep Cooperation Zone. Through its original Intuitive Neural Network brain-inspired algorithm, it successfully condenses supercomputing power into a device the size of a mini-fridge. While achieving a high degree of integration of supercomputing capabilities, it also provides a new path for improving the energy efficiency of traditional supercomputing centers and intelligent computing clusters.

Reporters learned that "Zhi Zhe No. 1" was jointly launched by two incubated companies of the Guangdong Academy of Intelligent Science and Technology—Zhuhai Hengqin Neo-Technology Technology Co., Ltd. and Suiren (Zhuhai) Medical Technology Co., Ltd. Within a very limited space, "Zhi Zhe No. 1" integrates 1152 CPU cores, 4.8TB of DDR5 memory, and 204TB of storage space. Its operating noise can be controlled below 45 decibels, and under maximum load, the device temperature can be stably controlled below 70 degrees Celsius.

“Traditional computing centers are like large buildings, requiring significant initial investment and high energy consumption; ‘Smart One’ is only the size of a mini single-door refrigerator, can operate directly from a household socket, and consumes only one-tenth the power of traditional supercomputing equipment,” said Nie Lei, director of the Joint Laboratory for Intelligent Computing Systems at the Guangdong Academy of Intelligent Science and Technology. He pointed out that this compact and efficient hardware design approach is not only suitable for standalone deployment scenarios, but also provides a new technical path for improving the energy efficiency density of existing computing facilities.

According to reports, the core breakthrough of "Zhi Zhe-1" lies in its unique Intuitive Neural Network brain-like algorithm. Its interpretable numerical and symbolic unified brain-like learning network highly simulates the human brain's computational mechanism, integrating symbolic computation and data-driven three-valued logic brain-like model, which can run various model training and inference, including large models, at high speed and efficiency.

In actual performance tests, "Zhi Zhe No. 1" performed outstandingly. On a single-node CPU, "Zhi Zhe No. 1" completed the training task of 10 billion tokens in just 30 hours, with training and inference speeds reaching 100,000 tokens/second and 500,000 tokens/second respectively, and its performance is comparable to traditional computing power clusters that require multiple high-end GPUs.

"The release of 'Smart Supercomputer-1' marks the beginning of a new era of 'mobile personal supercomputing' characterized by miniaturization, greening, personalization, accessibility, and mass adoption. It not only expands the deployment forms and applicable boundaries of supercomputing but also provides key technological support for building a greener, more flexible, and easier-to-use future computing power system," said Zhang Xu, Academician of the Chinese Academy of Sciences and President of the Guangdong Academy of Intelligent Science and Technology.