Recently, at the 2025 Artificial Intelligence Innovation Conference (HAIC2025), organized by the Photosynthetic Organization in Kunshan, Jiangsu Province, Sugon made its debut showcasing the real-world form of its scaleX ten-thousand-GPU supercluster. Based on the scaleX640 super-node system, the cluster has made remarkable strides in hardware performance, effectively tackling the prevalent issues of "fragmentation, lack of openness, and adaptability challenges" within China's domestic AI computing power ecosystem. This achievement signifies a new era of "ecosystem co-evolution" in the development of China's AI infrastructure.

The Sugon scaleX640 stands out with a computing power density per cabinet that is 20 times greater than that of traditional clusters, coupled with an impressively low Power Usage Effectiveness (PUE) value of 1.04. The scaleX ten-thousand-GPU supercluster is capable of deploying 10,240 AI acceleration cards, delivering a staggering total computing power of over 5 EFlops.

Its in-house-developed native RDMA high-speed network, ScaleFabric, ensures high communication efficiency and cost-effectiveness, facilitating the expansion of the cluster's scale. Moreover, Sugon has engineered a three-tier data transmission optimization system through thoughtful designs, enhancing AI inference performance and maximizing resource utilization. Furthermore, the integration of a digital twin and intelligent scheduling system for the supercluster guarantees high availability, ensuring the stable and efficient operation of large-scale AI tasks.