Amazon SageMaker Studio introduces GPU capacity reservation with flexible training plans
Amazon SageMaker Studio now supports GPU capacity reservations via Flexible Training Plans, offering cost savings and predictable access to resources. Users can manage their reservations through a self-service console.
Amazon SageMaker Studio Integrated Development Environments (IDEs), such as JupyterLab and the Code Editor, now offer GPU capacity reservations through SageMaker Flexible Training Plans (FTP). This feature provides users with predictable access to high-demand, high-performance computational resources while staying within their budget. By utilizing FTP, users can save up to 65% in costs compared to On-Demand instances when running machine learning workflows in JupyterLab or the Code Editor.
The FTP offers a completely self-service procurement experience. To begin, users should go to the SageMaker FTP console and choose their desired instance type, reservation duration, and start date for their Studio IDE workload. After reviewing and completing the purchase, users must wait for the plan to activate. When setting up a Studio app via the SageMaker Studio user interface, users can select their purchased plan from the Instance dropdown menu. SageMaker will automatically provision the instance without requiring any infrastructure management from the user.
As the plan approaches its expiration date, the IDE will proactively alert users, allowing them to save their work before the reservation concludes. For more information on using the FTP capacity reservation feature with Studio IDEs, users can refer to the documentation on using Training Plans with Studio IDEs. Additionally, details about launching JupyterLab and Code Editor applications in SageMaker Studio can be found in the Studio Spaces documentation.