SK텔레콤의 GPUaaS Full Stack 시스템 구현과 개발 방향

Learn about SK Telecom's comprehensive GPUaaS (GPU-as-a-Service) system in this technical presentation from the SK AI SUMMIT 2024. Explore how the full-stack solution enables AI service development, training, and inference by providing on-demand GPU resources and dedicated GPU allocations for extended periods. Discover the complete software stack required for GPUaaS implementation, from infrastructure to management, and understand the research and development directions for each component. Gain insights from SK Telecom's AI Cloud development team expert who specializes in cloud-native technologies, AI/ML infrastructure development, high-performance platform development for AI training and inference, and anomaly detection systems based on cloud metrics. Understand the optimization of Kubernetes clusters for AI workloads, MLOps pipeline construction, and the efficient management of AI model development, deployment, and monitoring processes, including the optimization of AI workloads using ARM architecture-based instances like AWS Graviton.