Databricks SDK for JavaScript
    Preparing search index...

    Interface ComputeSpec

    ComputeSpec: compute configuration — accelerator type and total accelerator count across all nodes.

    interface ComputeSpec {
        acceleratorCount?: number;
        acceleratorType?: ComputeSpec_AcceleratorType;
        provisionedCapacityId?: string;
    }
    Index

    Properties

    acceleratorCount? acceleratorType? provisionedCapacityId?

    Properties

    acceleratorCount?: number

    Total number of accelerators across all nodes. Must be a positive multiple of the per-node accelerator count encoded in accelerator_type. For example, GPU_8xH100 with accelerator_count: 16 allocates 2 nodes (8 GPUs per node).

    acceleratorType?: ComputeSpec_AcceleratorType

    Hardware accelerator type (for example, GPU_1xA10 or GPU_8xH100). The number of accelerators per node is encoded in the enum value — GPU_8xH100 means 8 H100 GPUs per node.

    provisionedCapacityId?: string

    Optional ID of a pre-provisioned accelerator capacity reservation to run this AI Runtime workload on. When set, the workload is scheduled onto the referenced reserved capacity instead of the on-demand capacity shared among all customers.

    Settings

    Member Visibility

    On This Page

    Properties