At the Baidu World Conference 2021 on August 18, Baidu announced that the second-generation Baidu Kunlun AI chip (the Kunlun Core 2 independently developed by Baidu), has entered the mass production phase.
The Kunlun Core 2 uses a 7nm process and is equipped with a self-developed second-generation XPU architecture. Compared to its first-generation predecessors, the performance is improved by 2–3 times. It is suitable for cloud, terminal, edge computing, and other scenarios. In the future, it will show its talents in fields such as autonomous driving, intelligent transportation, intelligent assistants, etc.
In addition to the self-developed XPU architecture and multiple independent designs, the Kunlun Core AI chip has also cooperated with many Chinese general-purpose processors such as FeiTeng, many Chinese operating systems such as Kirin, and Baidu’s self-developed flying paddle deep learning, completing the end-to-end framework adaptation and houses a full-stack Chinese AI capability that integrates software and hardware.
At the Baidu AI Developer Conference in 2018, Robin Li announced that Baidu will launch its self-developed AI chip Kunlun. The Kunlun chip is positioned as a general-purpose AI chip.
In December 2019, Baidu announced the completion of the research and development of the first Kunlun generation 1 AI chip for cloud computing and edge computing. According to Baidu, the Kunlun 1 can achieve 256TOPS INT8 processing capacity and 260 trillion operations per second using under 150 watts. Compared to the Nvidia V100S and the Cambrian Siyuan 270 launched at the same time, it houses superior computing power.
At the end of 2020, Baidu’s chief technology officer Haifeng Wang revealed that Kunlun 1 mass-produced over 20,000 units and achieved its large-scale deployment. The Kunlun 2 is expected to be mass-produced in the first half of 2021.