Making the Most Out of Hardware Accelerators in Kubernetes Clusters

Explore techniques for optimizing hardware accelerators in Kubernetes clusters for MLOps in this 25-minute conference talk. Learn how to efficiently manage GPUs and TPUs using open-source projects like Kubeflow, TensorFlow, and Kubernetes. Discover strategies for differentiating between compute instances and hardware accelerators, determining appropriate Pod placement, scaling, time-sharing accelerators, and establishing effective compute-accelerator links. Gain valuable insights to enhance machine learning workload performance and maximize the potential of your hardware resources in a Kubernetes environment.

Making the Most Out of Your Hardware Accelerators in a Kubernetes Clu... Rishit Dagli & Shivay Lamba