Today, Amazon Web Services announces three new features for Amazon Elastic Compute Cloud (Amazon EC2) Capacity Blocks for ML that enable you to get near-instantaneous access to GPU and ML chip instances through Capacity Blocks, extend the durations of your Capacity Blocks, and reserve Capacity Blocks for longer periods of up to six months. With these new features, you have more options to provision GPU and ML chip capacity to meet your machine learning (ML) workload needs.
With Capacity Blocks, you can reserve GPU and ML chip capacity in cluster sizes of one to 64 instances (512 GPUs, or 1,024 Trainium chips), giving you the flexibility to run a wide variety of ML workloads. Starting today, you can provision Capacity Blocks that begin in just minutes, enabling you to quickly access GPU and ML chip capacity. You can also extend your Capacity Block when your ML job takes longer than you anticipated, ensuring uninterrupted access to capacity. Finally, for projects that require GPU or ML chip capacity for longer durations, you can now provision Capacity Blocks for up to six months, allowing you to get capacity for just the amount of time you need.
EC2 Capacity Blocks are available for P5e, P5, P4d, and Trn1 instances in US East (N. Virginia and Ohio), US West (Oregon), Asia Pacific (Tokyo and Melbourne). See the User Guide for a detailed breakdown of instance availability by region.
To learn more, see the Amazon EC2 Capacity Blocks for ML User Guide.
Source:: Amazon AWS