Deploying Kubeflow to a Bare-Metal GPU Cluster from Scratch

My experience of deploying Google’s Kubernetes ML toolkit on physical servers with multiple GPUs

Attack on Kubeflow. Image by Anastasia Markovtseva, CC-BY-SA 4.0.

Hardware

I’ve got 3 standard Supermicro towers with 256GB RAM, an SSD, 5 HDDs, and 4 GPUs each. Ethernet connects them to the “controller” Dell server with access to the…