China's First Huawei Ascend 950DT Super-Node Cluster Goes Live in Wuhan
China's first 1024-card liquid-cooled super-node cluster powered by Huawei Ascend 950DT has officially gone live in Wuhan after a rapid one-month deployment. The cluster is designed to provide robust local computing power for large model training and AI agent inference. This launch marks a major milestone in China's domestic AI hardware ecosystem by addressing high-end computing supply shortages and delivering the massive scale required for trillion-parameter model training. It strengthens regional AI innovation and provides critical domestic infrastructure for advanced artificial intelligence workloads. The Ascend 950DT 1024-card liquid-cooled super-node utilizes full liquid cooling for its compute and interconnect cabinets, delivering an FP8 performance of 1 EFLOPS and 256TB of global unified memory. It features terabit-level inter-card bandwidth, microsecond-level end-to-end latency, and direct card-to-card accessibility via the Lingqu architecture.
## BACKGROUND
Huawei Ascend is a series of AI processors developed by Huawei for deep learning training and inference workloads. Technologies like the Lingqu architecture and advanced high-bandwidth memory systems are leveraged in these clusters to facilitate high-speed, low-latency communication across thousands of AI accelerator cards.