Source: H3C |Original Link
On July 18, under the guidance of the Organizing Committee Office of the World Artificial Intelligence Conference, and co-hosted by the Artificial Intelligence Industry Working Committee of the China Computer Industry Association, Shanghai Zhangjiang Hi-Tech Park Development Co., Ltd., and H3C Group, the "AI Computing Power Technology Architecture Innovation Forum" was successfully held at the Shanghai World Expo Center. The forum brought together authoritative standards organizations, leading domestic chip enterprises, top-tier internet companies, upstream and downstream industry chain partners, and industry user representatives to conduct in-depth exchanges on key topics including efficient collaboration of diverse computing power, open interconnection architectures, software-hardware collaborative optimization, and ecosystem adaptation. During the forum, the G-Link Universal High-Speed Interconnect Standard initiative was officially launched, advancing the standardization of open interconnection architectures and promoting the interoperability of domestic computing power.
Standards Drive Industrial Synergy, Advancing Open Interconnection
With the accelerated evolution of AI infrastructure, interoperability and ecosystem synergy among diverse computing powers have become critical factors influencing the release of computing efficiency. In response to this industrial trend, standards development has emerged as a vital pillar for driving technological innovation, industrial collaboration, and large-scale implementation. Hu Guodong, Deputy Secretary of the Party Committee of the China Electronics Information Industry Development Institute and President of the Artificial Intelligence Working Committee, stated in his address that artificial intelligence is a significant driving force behind the new round of technological revolution and industrial transformation. Facing the critical stage of rapid AI technological evolution and scaled industrial development, the industry must further consolidate consensus, adhere to demand-driven and application-oriented approaches with forward-looking planning, and solidly advance research and application verification of key standards for AI chips, thereby better supporting the high-quality development of China's AI chip sector and the broader AI industry.
Launch of G-Link Standards Initiative: Jointly Building a New Open Intelligent Computing Ecosystem
Currently, with the accelerated development of domestic diverse computing power, challenges such as disparate hardware architectures, non-unified interface standards, and fragmented software ecosystems have resulted in high costs from redundant development and adaptation across the industry chain, posing significant obstacles to the efficient collaboration of heterogeneous computing power. Addressing this industrial need, the forum officially launched the construction of the G-Link Universal High-Speed Interconnect Standard. Guided by the Artificial Intelligence Working Committee and jointly proposed by dozens of upstream and downstream industry chain partners including H3C, G-Link is an open high-speed interconnection architecture designed for AI high-performance computing and Scale-Up interconnection scenarios. It aims to reduce industrial adaptation costs through an open, unified interconnection architecture and standards, facilitate efficient interoperability and collaborative innovation among different chips and systems, and establish an open interconnection foundation for super-nodes and large-scale AI clusters. The launch of the G-Link standardization initiative marks its official opening to the industry, uniting industry chain partners to jointly advance standards development and provide new support for the efficient collaboration of diverse heterogeneous computing power and the construction of an open intelligent computing ecosystem. At the launch ceremony, H3C joined representatives from core AI industry chain partners—including standards organizations, research institutes, AI/GPU and interconnect chip enterprises, and computing power application providers—to witness this pivotal step in making the G-Link standard available to the industry.
Building High-Performance AI Infrastructure Through a Diverse and Open Ecosystem
As AI applications accelerate toward large-scale implementation, continuously improving overall computing system efficiency and optimizing cost-effectiveness per token have become critical industry imperatives. Liu Xinmin, Senior Vice President of H3C Group and Vice Chairman of the Technology Committee, delivered a keynote speech at the conference, sharing practices and insights on open ecosystems, heterogeneous computing synergy, and system-level efficiency enhancement. He noted that the focus of current AI competition has shifted from merely pursuing hardware scale to comprehensive competition encompassing system-level efficiency and ecosystem synergy. Adhering to the core philosophy of "Computing Power × Connectivity," H3C focuses on AI computing infrastructure, maintains openness and compatibility, leverages full-stack capabilities to support AI technological leaps, and helps the industry fully unlock the value of every unit of computing power.
In terms of specific implementation pathways, H3C upholds the philosophy of a diverse and open ecosystem, advancing coordinated efforts across multiple dimensions including heterogeneous computing, computing interconnection, integrated software, complete machine systems, and industrial models: by ensuring compatibility with mainstream xPU chips across multiple platforms to create a flexible and selectable heterogeneous computing foundation; by deeply integrating open interconnection protocols with computing networks to enhance the connection efficiency and collaborative capabilities of diverse computing power; by integrating and uniformly scheduling computing resources to break down resource silos and improve application adaptation and deployment efficiency; and by adopting standardized and modular designs for complete machine systems to drive AI infrastructure toward higher efficiency, greater reliability, and sustainable evolution. Looking ahead, H3C will join hands with industry chain partners to jointly build an open and collaborative next-generation AI infrastructure ecosystem.
Discussing Effective Computing Power Release and Exploring Key Pathways for an Open Intelligent Computing Ecosystem
During the subsequent roundtable dialogue, Li Fei, Vice President of H3C Group and General Manager of the Solution Development Department, engaged in discussions with experts from enterprises including Zhangjiang Haoheng, Tianshu Zhixin, Dongfang Suanxin, Alibaba Cloud, and ByteDance. They conducted in-depth discussions on core bottlenecks in computing power release, pathways for enhancing computing efficiency, and the construction of an open interconnection ecosystem. Drawing on H3C's practices in computing infrastructure, they shared perspectives on system-level synergy across dimensions such as complete machine systems, cluster architectures, interconnection capabilities, and ecosystem adaptation.
The key to realizing the true value of diverse computing power lies in open interconnection and efficient collaboration. Facing the new phase of large-scale AI application implementation, H3C will continue to collaborate with industry chain partners to promote the thriving development of an open intelligent computing ecosystem. Leveraging open, compatible infrastructure capabilities and collaborative adaptation capacities, H3C will continuously unleash the efficiency of diverse computing power, assisting industry customers in deploying AI applications with greater efficiency and optimized costs.