Huawei Technologies recently launched Flex:ai, an open-source software aimed at enhancing artificial intelligence (AI) chip utilization rates. This initiative reflects China’s ambition to develop top-tier AI training capabilities, even in the absence of advanced Nvidia processors. Built on Kubernetes, Flex:ai orchestrates a variety of processors, including graphics processing units (GPUs) and neural processing units (NPUs), optimizing their performance by allowing multiple AI workloads to operate simultaneously. Huawei claims this tool can boost processor utilization by an average of 30%. A smart scheduler, Hi Scheduler, manages the allocation of GPUs and NPUs, redistributing idle resources for enhanced efficiency. Kubernetes plays a crucial role as a backbone for AI infrastructure, facilitating the deployment and scaling of containerized applications. Zhou Yuefeng of Huawei emphasized that Flex:ai aims to democratize AI by maximizing compute resource usage while simplifying task management across diverse computing environments. This marks a significant step in fostering a self-sufficient AI ecosystem in China.
Source link
Share
Read more