GPU Sharing and Container Device Interface in Kubernetes Device Plugins
CNCF [Cloud Native Computing Foundation] via YouTube
Overview
Explore GPU sharing and Container Device Interface (CDI) in Kubernetes Device Plugins through this informative conference talk. Dive deep into efficient management of compute resources, particularly GPUs, for AI/ML workloads running on Kubernetes. Learn about the implementation of features in k8s Device Plugins that make resources accessible to end-user applications. Discover the importance of partition and resource sharing for improved utilization and cost reduction. Examine the Container Device Interface (CDI) as a new option for Device Plugin authors and the flexibility it provides, including resource sharing for GPUs. Starting with use cases, investigate how a Device Plugin exposes GPUs and various sharing options to enhance device utilization and right-sizing for workloads, such as time slicing, MIG, and MPS. Gain insights into Kubernetes integration with devices and CDI, GPU sharing mechanisms, and how applications and frameworks can leverage this functionality.
Syllabus
Sharing Is Caring: GPU Sharing and CDI in Device Plugins - Evan Lezar, NVIDIA & David Porter, Google
Taught by
CNCF [Cloud Native Computing Foundation]