Recently, the inaugural ‘Zhenwu’ 10,000-GPU intelligent computing cluster in the Guangdong-Hong Kong-Macao Greater Bay Area commenced operations at the Shaoguan Data Center. This marks the first project to be realized in the Greater Bay Area as part of the national new infrastructure initiative focused on ‘ultra-large-scale intelligent computing clusters.’ The cluster embodies full-chain independent research and development, signifying China’s strides in deploying cutting-edge computing capabilities on an industrial scale.

The Zhenwu cluster has achieved multiple technological breakthroughs. Notably, it boasts an end-to-end network latency of a mere 4 microseconds and a network peak utilization rate surpassing 95%. These features cater to the communication demands of large-scale model training and facilitate tasks involving models with hundreds of billions of parameters. The comprehensive Zhenwu chip system is characterized by its expansive memory and high interconnection bandwidth, offering robust support for the domestic computing power infrastructure.