The National Research Platform (NRP) gives researchers access to a distributed, Kubernetes-based cyberinfrastructure spanning dozens of institutions, including a large pool of GPUs and high-throughput storage. For researchers coming from traditional HPC batch schedulers, the model can be unfamiliar: the unit of work is a containerized workload rather than a script submitted to a queue.

In this interactive workshop, participants will learn how to run tasks on the platform. We start with the conceptual model — namespaces, Pods, containers, and how scheduling differs from a queue — then move directly to practice. Participants will launch an interactive Pod for development, submit a batch Job, and request a specific GPU type, learning how resource requests and limits affect placement. We cover attaching a persistent volume so results survive the Pod, and close with reading logs, diagnosing common failure modes, and norms for sharing a multi-institution resource.

Prerequisites: participants should arrive with NRP account access and kubectl already installed. Setup instructions will be circulated in advance, and help will be available before the start. Users unfamiliar with containerization are recommended to attend the “Containerization for Reproducible Scientific Computing” workshop.