Content |
History
2025: System Launch
On December 3, 2025, China launched the world's largest distributed computing infrastructure for artificial intelligence. A system called Future Network Test Facility (FNTF) through a high-speed optical network combines data centers located in various parts of the country.
According to the South China Morning Post, the project took more than ten years to implement. The infrastructure includes data centers in 40 Chinese cities. The total length of the optical network exceeds 55 thousand km. It is argued that the distributed system works almost as quickly as a single supercomputer: the declared efficiency reaches 98% of the indicators of individual data centers of a comparable class. A set of advanced technologies is involved to provide wide bandwidth with guarantees of ultra-low latency and almost zero packet loss.
FNTF delivers the highest performance. An example is the training of a large model with hundreds of billions of parameters. Such a process usually requires more than 500 thousand iterations. On the new network, each iteration takes about 16 seconds. In conventional HPC systems, the figure can increase by more than 20 seconds. Thus, in general, FNTF in AI training provides time savings of several months.
The platform can simultaneously support 128 heterogeneous networks and perform 4096 test loads. The FNTF infrastructure opens up a whole new opportunity for applications with high real-time processing requirements, such as telemedicine and industrial Internet. In the future, the system is planned to be used in the implementation of complex and resource-intensive projects in the scientific field.[1]
