Managing Compute Resources on the H2O Platform
- H2O AI Managed Cloud
- H2O MLOps
- H2O Driverless AI
- Enterprise h2oGPTe
This lesson walks through how H2O orchestrates AI workloads on Kubernetes as an AI program scales across teams. Driverless AI experiments, Feature Store operations, MLOps deployments, and h2oGPTe agent operations all run as managed Kubernetes workloads, with administrators defining resource profiles, such as a large training profile for AutoML jobs or a GPU inference profile for computer vision. Data scientists then pick from the profiles available to them without needing Kubernetes expertise. The lesson also covers cost controls, including idle timeouts and cluster autoscaling.
Key takeaways
- How H2O runs Driverless AI, Feature Store, MLOps, and h2oGPTe workloads as managed Kubernetes jobs
- The role of admin-defined resource profiles, such as a large training profile or a GPU inference profile, in allocating CPU, GPU, and memory
- How data scientists select a resource profile without needing Kubernetes expertise themselves
- What cost guardrails like idle timeouts and maximum run durations do to prevent stalled or runaway jobs
- How cluster autoscaling adds or removes nodes automatically as workload demand rises and falls
Start owning your AI