View a markdown version of this page

AcceleratorPartitionLayout - Amazon SageMaker

AcceleratorPartitionLayout

A layout of accelerator partitions for a production variant. On NVIDIA GPUs, the partitions are Multi-Instance GPU (MIG) instances. SageMaker applies the layout to every accelerator on the variant's instance type. You can't change the layout of a running endpoint. To use a different layout, create a new endpoint.

Contents

AcceleratorPartitionTypes

The list of partition profiles to carve on each accelerator that this layout applies to. On NVIDIA GPUs these are Multi-Instance GPU (MIG) profiles, for example ["mig-3g.40gb", "mig-4g.40gb"]. The order of the profiles doesn't change how the accelerator is partitioned. However, when you update an endpoint, a layout with the same profiles in a different order counts as a different layout, and the update is rejected. Repeat a profile to request more than one partition of that size: for example, ["mig-3g.40gb", "mig-3g.40gb"] requests two mig-3g.40gb partitions.

Type: Array of strings

Array Members: Minimum number of 0 items. Maximum number of 32 items.

Length Constraints: Minimum length of 0. Maximum length of 32.

Pattern: mig-[0-9]+g\.[0-9]+gb([+-][a-z0-9.]+)*

Required: No

See Also

For more information about using this API in one of the language-specific AWS SDKs, see the following: