how it will work
How it fits into GPU.ai
Ray Clustering is a planned way to run a Ray cluster on GPUs you rent through GPU.ai. You will pick the GPU type and the number of nodes, and GPU.ai will launch the machines, connect them into one cluster, and hand you an address to submit Ray jobs to.
- Your Ray programsubmitted to one address
- Ray head nodeschedules tasks
- GPU nodesrented on GPU.ai
Pick a shape
Choose a GPU type and a node count. GPU.ai launches the machines and forms the cluster.
Your Ray code as it is
Programs written for Ray are meant to run the same way they do on any other Ray cluster.
Shut down when done
Add or remove GPU nodes as work changes, and stop the cluster when the job ends so idle GPUs stop billing.