Alibaba Cloud teases new modular datacenter architecture

Alibaba Cloud has unveiled its CUBE DC 5.0 modular datacenter architecture at the annual Apsara conference, aiming to meet the increasing demand for AI infrastructure. The architecture utilizes prefabricated modular designs and advanced proprietary technologies, including a wind-liquid hybrid cooling system, all-direct current power distribution, and a smart management system. The modular approach reportedly reduces deployment times by up to 50% compared to traditional datacenter building methods. The company has not provided detailed explanations of the technologies, but the Register has inquired about the wind-liquid hybrid cooling system, which, when translated into Mandarin and searched, refers to cold plate cooling. This technique involves circulating cooled liquids on hardware to achieve cooling. CEO Eddie Wu emphasized Alibaba Cloud s commitment to building an AI infrastructure for the future, with a focus on efficiency and sustainability. The company has also introduced a scheduler to better manage hardware resources, achieving up to 90% utilization rates. Alibaba Cloud s Enterprise Elastic Compute Service (ECS) has reached its ninth generation, with improved recommendation engine speeds and database query performance. The company also announced the Open Lake data utility, which integrates multiple big data engines for generative AI applications. Additionally, Alibaba Cloud introduced DMS: OneMeta+OneOps, a new offering that manages metadata from 40 different data sources. In 2024, the company released its Qwen 2.5 multimodal models, supporting 29 languages and offering enhanced knowledge and stronger capabilities in math and coding. Lastly, Alibaba Cloud introduced Tongyi Wanxiang, a text-to-video AI model that works with both Chinese and English prompts, capable of generating high-quality videos in various visual styles. Overall, Alibaba Cloud s new architecture and AI advancements aim to set global standards for efficiency and sustainability in AI infrastructure.

Source: theregister.com
Published on 2024-09-20