Last week, China’s leading tech firms unveiled a new generation of chips designed for ‘Global Scheduling Ethernet’ (GSE), a networking protocol optimized for the intense demands of AI and other high-performance computing workloads. China Mobile has spearheaded this development, releasing a detailed white paper outlining the GSE Ethernet’s technical framework, echoing the motivations of the Ultra Ethernet Consortium (UEC).
The UEC, which includes industry giants like Intel and AMD, aims to tailor Ethernet protocols to better suit AI and HPC applications that require low-latency solutions over extensive networks. High traffic volumes and vast network sizes have rendered traditional Ethernet inadequate, prompting the need for these specialized adaptations.
China Mobile’s strategy involves innovative techniques such as fixed-size packet containers and a Dynamic Global Scheduling Queue to create streamlined, efficient network pathways.
Reports from Chinese media highlight that over 50 companies, including chip makers and universities, collaborated on this technology, which has already demonstrated significant performance enhancements in AI training at scale.
While remaining cautious, experts suggest that if GSE outperforms existing consortium solutions, it might provoke political and economic challenges, potentially leading to restricted international technology transfers. This could also mean that UEC members may face additional obstacles when approaching markets led by Chinese technologies.
Efforts to reach out to China Mobile for comments have not yielded responses, and it remains to be seen how this development will impact the global networking landscape.