SageMaker AI Launches SOCI Lazy-Loading Images to Speed Up Container Startup
Understand the story
On October 28, 2026, AWS announced on its Machine Learning Blog (primary source) that Amazon SageMaker AI has introduced Seekable OCI (SOCI) index images for training and processing jobs. The technology uses lazy loading: when a container starts, it only pulls the files it needs, with the rest of the data streamed in the background. According to AWS's measured results, download time for 1.8–35.4 GB images dropped from up to 7 minutes 17 seconds to 3–11 seconds, roughly an 8x to 46x improvement; GPU idle time was cut by up to 90%. The trade-off is an extra 3–16 seconds for the first epoch, but overall each job finishes 1–6 minutes earlier. The feature is now available for SageMaker AI training and processing jobs, with no further reports since.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- AWS Machine Learning BlogAnnouncing SOCI-indexed images for faster SageMaker AI jobs
AWS has introduced Seekable OCI (SOCI) indexed images for SageMaker AI training and processing jobs, using lazy loading to fetch only the files needed to start the container while the rest streams in the background. In tests, downloads of 1.8–35.4 GB images dropped from up to 7 minutes 17 seconds to 3–11 seconds (about 8x–46x), GPU idle time was cut by up to 90%, and while the first epoch took 3–16 seconds longer, each job overall finished 1–6 minutes earlier.
Story heat trend
Not enough continuous observations to show a trend yet.