According to the Caixin news agency, Baidu Group Executive Vice President Shen Jie revealed that the core of the AI IaaS layer of Baidu Cloud Intelligence 3.0, Kunlun core has been mass-produced tens of thousands of pieces, to achieve large-scale commercialization of landing, Kunlun core 3 generations will be mass production in early 2024.
Last August, Baidu announced the mass production of its second-generation Baidu Kunlun AI chip, Kunlun Core 2. The chip adopts 7nm process and is equipped with self-developed second-generation XPU architecture, which improves performance by 2-3 times compared to the first generation and is suitable for multiple scenarios such as cloud, end and edge.
Last September, Baidu Kunlun core business analyst Song Chunxiao said, Kunlun core is the only domestic chip that has been refined by the Internet’s large-scale core algorithm, landing more than 20,000 pieces.
The Kunlunxin AI chip not only has its own XPU architecture and a number of independent designs but also has completed end-to-end adaptations with a number of domestic general-purpose processors such as Feiteng, a number of domestic operating systems such as Kirin and Baidu’s self-developed Flying Paddle deep learning framework, with full-stack domestic AI capabilities of software and hardware.
