Gaudi-based Amazon EC2 DL1 Training Instances

Learn to how set up and run Habana-based Amazon EC2 DL1 Training Instances

Habana Gaudi-based Amazon EC2 DL1 Training Instances feature up to 8 Gaudi accelerators and deliver up to 40% better price performance than current generation GPU-based EC2 instances. These new high-efficiency instances feature 32GB of high bandwidth memory (HBM) per accelerator, 768 GiB of system memory, custom 2^nd Generation Intel® Xeon® Scalable Processors, and 4 TB of local NVMe storage. In addition, DL1 instances provide 400 Gbps of networking throughput and feature all-to-all connectivity within DL1 servers with Gaudi’s native integration of ten 100 Gigabit ports of RDMA over Converged Ethernet.

Habana’s SynapseAI® softwaren is optimized for Gaudi accelerators and natively integrates TensorFlow and PyTorch frameworks on popular AI models optimized for computer vision and natural language processing applications.  

You can build new or migrate existing models with Habana’s programmable architecture, designed expressly for developer ease of use and flexibility. You can get started on DL1 instances, using Amazon ECS and Amazon EKS for containerized applications, or with Amazon SageMaker–a managed service for building, training and deploying machine learning applications. Get instant access to Gaudi accelerators and build or migrate your existing models with documentation, how to content and tools found here and on Habana’s GitHub.  

You can use the links below to get started with setting up a Gaudi-based instance and on EC2. You can begin with a pre-built AWS Deep Learning AMI (DLAMI) or AWS Deep Learning Container (DLC) when launching an instance from EC2, or use a Base AMI or future Images from the AWS marketplace.