Why compute matters
Data science and machine learning workloads have widely varying needs. Some run on a single small instance; others need many machines at once, or specialized hardware such as GPUs. Rather than provisioning and managing that infrastructure yourself, you define the compute that should be available and let the platform provision it on demand and scale it as workloads require.How compute is organized
Compute is grouped into pools. Each pool defines the instance types it provides and the bounds within which it scales, and an administrator makes pools available to specific perimeters. Because a workload runs in a perimeter, it can only use the compute assigned to that perimeter. This keeps one team’s compute separate from another’s and lets administrators direct different kinds of work to appropriate hardware. Compute on the platform can come from more than one source:- Cloud compute
- Capacity provisioned in your own cloud account, in the same cloud as your data plane.
- External compute
- Capacity from outside your primary cloud, including on-premises clusters and additional providers, connected into your data plane so it appears alongside your cloud compute.