Amazon EC2 Capacity Blocks now supports instant start times and extensions

Amazon Web Services has announced three new features for EC2 Capacity Blocks for ML, designed to improve GPU and ML chip instance provisioning:

• Near-instantaneous capacity access for ML workloads
• Ability to extend Capacity Block durations during ongoing jobs
• Option to reserve GPU/ML chip capacity for up to six months

Key capabilities include:

• Reserve 1-64 instances (up to 512 GPUs or 1,024 Trainium chips)
• Support for P5e, P5, P4d, and Trn1 instances
• Available in multiple US and Asia Pacific regions

These enhancements provide more flexibility and responsiveness for machine learning workload capacity management.