SageMaker Studio GA: EC2 P4de in Tokyo, Singapore, Frankfurt with 640GB GPU memory and lower training costs

AI Impact Summary

Amazon EC2 P4de instances (8 NVIDIA A100 GPUs with 80GB each, 640GB total) are GA in Tokyo, Singapore, and Frankfurt on SageMaker Studio notebooks. The significantly larger GPU memory and 60% faster training potential, with an estimated 20% lower cost versus P4d, will shorten training cycles for large, high-resolution datasets and reduce total training spend for frequent retraining workflows. This regional expansion enables in-region workloads with SageMaker Studio notebooks, reducing data transfer latency and improving data residency, while providing a clear path for migrating existing P4d-based workloads. Refer to developer guides to configure JupyterLab and CodeEditor in SageMaker Studio and review pricing for TCO impact.