Announcing preview of SageMaker Model Training support for ml.p4de instances
Amazon SageMaker model training now supports preview of ml.p4de.24xlarge(p4de) instances, powered by NVIDIA A100 GPUs and high-performance HBM2e GPU memory. These instances provide the highest performing compute infrastructure currently available for use cases that require training of large language models (LLMs), computer vision (object detection, semantic segmentation) and generative artificial intelligence.
Amazon SageMaker model training now supports preview of ml.p4de.24xlarge(p4de) instances, powered by NVIDIA A100 GPUs and high-performance HBM2e GPU memory. These instances provide the highest performing compute infrastructure currently available for use cases that require training of large language models (LLMs), computer vision (object detection, semantic segmentation) and generative artificial intelligence.
P4de instances feature up to 8 NVIDIA A100 GPUs with 80GB high-performance HBM2e GPU memory, 2X higher than the GPUs in p4d instances. The higher GPU memory primarily helps reduce the training time and cost for workflows involving large datasets of high-resolution data and complex models. Similar to p4d, they also support up to 96 vCPUs, up to 400 Gbps of network bandwidth with EFA and GPUDirect RDMA, and up to 8 TB local NVMe-based SSD storage.
SageMaker Model Training supports preview of p4de instances in the AWS US East (N. Virginia) and US West (Oregon) regions.
To learn more about p4de instances, visit the AWS news blog or the Amazon EC2 P4 instance page. To get started with using p4de instances read the Amazon SageMaker documentation and the SageMaker Model Training page. For more information about p4de pricing visit Amazon SageMaker pricing web page.