Minimizing GPU Cost for Deep Learning on Kubernetes

Explore a GPU sharing solution for native Kubernetes to minimize costs and improve efficiency in deep learning tasks. Learn how to define GPU sharing API, implement scheduling without modifying core scheduler code, and integrate GPU isolation with Kubernetes. Discover techniques to run multiple TensorFlow jobs on a single GPU device within a Kubernetes cluster, significantly enhancing GPU usage for AI model development, debugging, and inference services. Gain insights from Alibaba experts on addressing the challenge of wasted GPU resources in clusters and optimizing deep learning workflows on Kubernetes.

Minimizing GPU Cost for Your Deep Learning on Kubernetes - Kai Zhang & Yang Che, Alibaba