On August 31, the 18th China IDC Industry (Yangtze River Delta) Annual Ceremony with the theme of "AI Creation and Computing Leads to Change" was grandly held in Shanghai today. The conference is jointly sponsored by the China IDC Industry Annual Ceremony Organizing Committee, the "Computing Power Pujiang" Special Committee of the Shanghai Communications Society, and the Data Center Committee of the China Communications Industry Association. Cao Hui, deputy general manager of the Product Operations Department of Zhongke Yushu, was invited to deliver a keynote speech at the main venue - "Efficient AI Large Model Computing Power Base Based on DPU".
▲ Cao Hui, deputy general manager of the Product Operations Department of Zhongke Yushu, was invited to deliver a keynote speech
With the continuous development of artificial intelligence, traditional computing power base has become a bottleneck restricting the improvement of AI computing power. Cao Hui emphasized in his speech that the thirst for computing power in large-scale AI model applications represented by ChatGPT is driving the evolution of computing power bases in the direction of large bandwidth and low latency. In this context, the emergence of data processing units (DPUs) provides key support for the upgrade of AI computing power bases, providing it with the necessary large bandwidth and low latency capabilities.
Cao Hui mentioned that the three-in-one U (CPU/GPU/DPU) has become the new computing power base of the intelligent computing center, which can realize the hardware collaborative integration of general computing power, intelligent computing power, and basic computing power, so that the intelligent computing center can not only It can cope with complex and diverse tasks and provide solutions to increasingly challenging AI application requirements.
Cao Hui also mentioned "Iaas on DPU", which is the integration of acceleration of various infrastructure layers in DPU, including computing acceleration, storage acceleration, network acceleration, security acceleration, cloud native acceleration, etc. The company has also formed a rich The product matrix layout can not only help improve the overall performance of the AI computing power base, but also provide users with more efficient and complete infrastructure solutions.
Zhongke Yushu will continue to uphold the concepts of technological innovation and open cooperation, and welcomes upstream and downstream partners such as server manufacturers, CPU/GPU manufacturers, operating system manufacturers, etc. to join the Yushu DPU ecosystem to jointly promote the development of AI computing power base and provide AI Make positive contributions to industrial prosperity.