Pandaily reported on 18 October that Peng Cheng Laboratory had begun trial operation of 10,000P of domestically built AI computing on Cloud Brain III in Shenzhen. The milestone was announced at the opening of the World Computing Network Conference, held in the city’s Nanshan district. The laboratory says the capacity online is supporting domestic large-model training and a research network for AI for science, while construction continues towards a planned 16,000P.
Computing capacity measured in petaflops describes how many calculations a system can perform each second. The laboratory describes Cloud Brain III as using a domestic supernode architecture intended to ease data-transfer bottlenecks. Pandaily reported that IT Home identified Huawei’s Ascend 384 SuperPoD architecture, while the laboratory’s announcement did not name a chip supplier.
Writing a work report could eventually involve a model trained on this capacity, if such a model were made available for that task. For now, the uses the laboratory describes are model training and scientific research.
The conference also released two white papers, including one produced with Huairou Laboratory on coordinating computing and power systems. Its roadmap runs through 2035.
