What changed
AWS introduced SOCI-indexed images for Amazon SageMaker AI Training and Processing jobs, which significantly reduces container image download time. This technology, called Seekable OCI (SOCI), allows SageMaker to lazy-load container images, fetching only the necessary files to start the container immediately and streaming the rest in the background.
What this means in practice
With SOCI indexing, SageMaker AI jobs can start in seconds instead of minutes, especially for large deep learning images that often exceed 10 GB. The reduction in download time translates to lower job costs and more time spent on model development rather than waiting for infrastructure. The feature is available by pointing to a pre-indexed AWS Deep Learning Container (DLC) or adding an index to a custom image, without requiring changes to job definitions, entry points, or data channels.
Limitations and availability
SOCI-indexed images are available for use with Amazon SageMaker AI Training and Processing jobs. The feature automatically checks for an index when creating a job, and if it is not found, it falls back to a standard download. The performance improvements were measured across eight image sizes, with download times reduced by up to 38x for the largest image. Training performance is unaffected beyond a first-epoch penalty, which is consistent across instance types and sizes. The results were measured on specific instance types and regions, and the feature is expected to be available for all supported instances.
