Recently, at the main forum of the 2025 China Computing Power Conference, China Mobile, together with dozens of industrial chain partners including Zhejiang Lab and Moore Threads, jointly launched the strategic cooperation for the co-construction of the Open Intelligent Computing Interconnection (OISA) ecosystem, and released the OISA 2.0 protocol. This initiative marks a solid step forward for China’s intelligent computing industry in the fields of technological innovation and ecological collaboration, realizing the capability aggregation and upgrading of the computing power industrial chain.
As large models evolve toward higher parameters and more modalities, ultra-large-scale clusters composed of tens of thousands of AI chips have become a global development trend in the intelligent computing industry. High-speed interconnection between chips, serving as the "central nervous system" of clusters, directly determines the efficiency and scalability of the entire intelligent computing system, and is a key bottleneck for in-depth breakthroughs in AI technology.
The OISA (Omni-directional Intelligent Sensing Interconnection Architecture) protocol standard aims to build an efficient, intelligent, flexible, and open AI chip interconnection system, with next-generation advanced interconnection capabilities featuring all-round connectivity and intelligent sensing. By enabling efficient interconnection, this system can support the vertical expansion of the performance of super-node intelligent computing clusters. Compared with the OISA 1.1 version, OISA 2.0 further increases the number of supported AI chips to 1,024, breaks through the TB/s level in bandwidth, and shortens the AI chip interconnection latency to hundreds of nanoseconds, providing strong support for data-intensive AI applications such as large-model training, inference, and high-performance computing.
As a domestic GPU innovation enterprise invested by China Mobile and an ecological partner, Moore Threads has deeply participated in the formulation of the OISA standard, actively contributing its core technological capabilities in the fields of AI and computing, and promoting the construction of an open and collaborative domestic software and hardware ecosystem. Currently, the company’s flagship AI training and inference products fully support the OISA 1.1 standard, demonstrating significant advantages in compatibility, collaboration, and performance, and achieving extensive adaptation and efficient support for mainstream AI frameworks, models, and applications.
At the critical stage of OISA 2.0 implementation, Moore Threads continues to work closely with industry partners to jointly advance the evolution and implementation of this standard. In subsequent product iterations, Moore Threads will continuously integrate the latest technical requirements of OISA, constantly enhance product advancement and optimize the architecture, helping China’s GPU basic software and hardware ecosystem achieve a higher level of independent innovation and integrated development.
(Reprinted from China Grid https://news.eccn.com)