Core Insights - Huawei has officially launched a groundbreaking AI technology called Flex:ai, which can enhance the average utilization rate of computing resources by 30% in AI workloads where full card computing power is not utilized [1] Group 1: Flex:ai Overview - Flex:ai is a software for pooling and scheduling various types of processors (XPU) built on Kubernetes, similar to Nvidia's Run:ai technology, but with two unique advantages [1] - The technology is positioned as a crucial tool for the industrialization of AI, with Huawei planning to open-source it to the Magic Engine community to build a complete ModelEngine open-source ecosystem [1] Group 2: Key Capabilities of Flex:ai - Flex:ai addresses the need for AI container technology, which is essential in the era of large models, by providing lightweight virtualization that allows seamless migration of model code and runtime environments [2] - The technology can dynamically mount GPU and NPU resources, enhancing overall resource utilization in clusters [2] - It is projected that by 2027, over 75% of AI workloads will be deployed and run using container technology [2] Group 3: Resource Management and Scheduling - Flex:ai can partition a single GPU/NPU card into multiple virtual computing units with a precision of 10%, allowing multiple AI workloads to run simultaneously on a single card [3] - The technology features unique virtualization capabilities, including "remote virtualization," which aggregates idle XPU resources across the cluster into a "shared computing pool" [5] - Flex:ai's intelligent scheduling can automatically sense cluster load and resource status, optimizing resource allocation based on workload priority and requirements [6] Group 4: ModelEngine Open-source Ecosystem - The release and open-sourcing of Flex:ai will support Huawei in building a comprehensive ModelEngine open-source ecosystem, which includes previously released tools like Nexent, AppEngine, DataMate, and UCM [7] - ModelEngine serves as an AI platform for model training, inference, and application development, providing a one-stop service from data preprocessing to model deployment [7] Group 5: Industry Application and Collaboration - Flex:ai is designed to facilitate the practical application of AI container technology across various industries, addressing the diverse needs of AI workloads [9] - The open-source nature of Flex:ai encourages collaboration among academia, industry, and developers to establish standards for heterogeneous computing virtualization and AI application integration [9]
见证历史!华为,重大发布
Zhong Guo Ji Jin Bao·2025-11-21 09:42