| Resource | Minimum | Notes |
|---|---|---|
| Worker nodes | 2 | Minimum distributed cluster footprint |
| CPU per node | 2 cores | Applies to head and workers |
| Memory per node | 4 GB | Applies to head and workers |
If GPU count is greater than zero for the head or any worker node, OPF-Ray uses a GPU-enabled image for the cluster. For CPU-only workloads set GPU count to 0.
Important: Ray Train and Tune workflows that run across multiple nodes require shared persistent storage for checkpoints. Depending on the deployment environment, this may be Amazon S3, Azure Blob Storage, or a shared/distributed file system.