GPU Sharing and Container Device Interface in Kubernetes Device Plugins

Explore GPU sharing mechanisms and Container Device Interface (CDI) implementation in Kubernetes Device Plugins through this technical conference talk. Gain deep insights into managing AL/ML workloads on Kubernetes with efficient GPU resource allocation, focusing on partition and resource sharing strategies for cost optimization. Learn about various GPU sharing options including time slicing, MIG, and MPS, while understanding how Kubernetes integrates with devices and CDI. Discover how Device Plugin authors can leverage CDI's flexibility to expose GPUs and implement different sharing options for improved device utilization. Master the techniques for right-sizing workloads and integrating applications and frameworks with these functionalities, presented by experts Christopher Desiniotis from NVIDIA and David Porter from Google.