고성능, 고효율 AI Computing 환경을 위한 차세대 GPU Fabric 및 NVLink 가상화 기술

Learn about advanced GPU infrastructure technologies in this Korean conference talk from SK AI SUMMIT 2024. Discover how to build high-performance, efficient GPU clusters for on-demand AI infrastructure utilization by multiple customers. Explore the fundamentals of NVLink and RDMA technologies that maximize data transfer speeds between GPUs, which is crucial as AI models become increasingly complex. Understand virtualization techniques that enable dynamic and efficient sharing of GPU resources among multiple users. Gain insights from the speakers' experiences in developing cost-effective solutions for expensive GPU utilization while maintaining flexibility and scalability. The presenters, Taekyung Kim and Geon Lee from SK Telecom, share their expertise in cloud network infrastructure and software-defined networking, discussing their work on Petasus Cloud virtualization solution and their vision for creating resilient, high-performance infrastructure software.