Search
The Shuguang 8000 (Dengfeng) marks China’s inaugural fully domestically developed 100,000-GPU AI supercomputing cluster, leveraging an advanced immersion phase-change liquid cooling system to effectively overcome the thermal management hurdles inherent in ultra-high-performance computing environments. As the deployment of AI accelerator cards scales up to 100,000 units, the power density of the chips rises dramatically, straining the capabilities of traditional air cooling and cold plate cooling solutions to their breaking points. Elevated temperatures can trigger chip frequency throttling and compromise the overall stability of the cluster, underscoring the necessity of liquid cooling technology in fully harnessing the potential of China’s high-performance chips.
-
C114 Communication Network -
Communication Home
