China Telecom's Two 10,000-Card Clusters Put into Production and Computing Power Developments

TapTechNews August 9th news, according to Daily Economic News report, The two 10,000-card clusters of China Telecom in Shanghai and Beijing have been put into production currently.

In addition, China Telecom uses the intelligent computing network and computing power scheduling platform to realize the interconnection and on-demand invocation of national computing power resources. The company's self-developed Xirang computing power scheduling platform has achieved a national computing power scheduling of 21 EFlops.

Shao Guanglu, the director of the Science and Technology Committee of China Telecom, said that in the future, the intelligent computing cluster for AI large model training needs to upgrade from IDC (TapTechNews note: Internet Data Center) to AIDC (AI-driven Data Center). Focusing on indicators such as model computing power utilization rate MFU, linear acceleration ratio, stability, latency, packet loss rate, and GPU utilization rate, continuous work such as the design, construction, operation, and optimization of the intelligent computing cluster is carried out to improve the efficiency and effectiveness of large model training.

Shao Guanglu also proposed that the networking of computing power has become the development direction, suggesting promoting the open source and openness of the computing power ecosystem, providing a basic development environment for heterogeneous computing power intercommunication, connecting various heterogeneous hardware with a unified standard to facilitate the free switching of large models among heterogeneous computing powers, and at the same time promoting co-construction and sharing.