AWS Announces GA of Habana Labs-powered EC2 DL1 Instances

Print Friendly, PDF & Email

HPC ClustersSEATTLE– Oct. 26, 2021– AWS has announced general availability of Amazon Elastic Compute Cloud (Amazon EC2) DL1 instances designed for training machine learning models and powered by Gaudi accelerators from Habana Labs (an Intel company). Providing up to 40 percent better price performance for training machine learning models than the latest GPU-powered AWS EC2 instances, according to Amazon, the DL1 instances are intended for machine learning models for use cases like natural language processing, object detection and classification, fraud detection, recommendation engines, intelligent document processing and business forecasting, and more.

Amazon said DL1 instances are available on demand via a low-cost pay-as-you-go usage model with no upfront commitments.

New DL1 instances use Gaudi accelerators built specifically to accelerate machine learning model training by delivering higher compute efficiency at a lower cost compared to general purpose GPUs, according to Amazon. DL1 instances feature up to eight Gaudi accelerators, 256 GB of high-bandwidth memory, 768 GB of system memory, 2nd generation Amazon custom Intel Xeon Scalable (Cascade Lake) processors, 400 Gbps of networking throughput, and up to 4 TB of local NVMe storage. Together, these innovations translate to up to 40% better price performance than the latest GPU-powered Amazon EC2 instances for training common machine learning models. Customers can quickly and easily get started with DL1 instances using the included Habana SynapseAI SDK, which is integrated with leading machine learning frameworks (e.g. TensorFlow and PyTorch), helping customers to seamlessly migrate their existing machine learning models currently running on GPU-based or CPU-based instances onto DL1 instances, with minimal code changes. Developers and data scientists can also start with reference models optimized for Gaudi accelerators available in Habana’s GitHub repository, which includes popular models for diverse applications, including image classification, object detection, natural language processing, and recommendation systems.

“The use of machine learning has skyrocketed. One of the challenges with training machine learning models, however, is that it is computationally intensive and can get expensive as customers refine and retrain their models,” said David Brown, Vice President, of Amazon EC2, at AWS. “AWS already has the broadest choice of powerful compute for any machine learning project or application. The addition of DL1 instances featuring Gaudi accelerators provides the most cost-effective alternative to GPU-based instances in the cloud to date. Their optimal combination of price and performance makes it possible for customers to reduce the cost to train, train more models, and innovate faster.”

Customers can launch DL1 instances using AWS Deep Learning AMIs or using Amazon Elastic Kubernetes Service (Amazon EKS) or Amazon Elastic Container Service (Amazon ECS) for containerized applications. For a more managed experience, customers can access DL1 instances through Amazon SageMaker, making it even easier and faster for developers and data scientists to build, train, and deploy machine learning models in the cloud and at the edge. DL1 instances benefit from the AWS Nitro System, a collection of building blocks that offload many of the traditional virtualization functions to dedicated hardware and software to deliver high performance, high availability, and high security while also reducing virtualization overhead. DL1 instances are available for purchase as On-Demand Instances, with Savings Plans, as Reserved Instances, or as Spot Instances. DL1 instances are currently available in the US East (N. Virginia) and US West (Oregon) AWS Regions.

Riskfuel provides real-time valuations and risk sensitivities to companies managing financial portfolios, helping them increase trading accuracy and performance. “Two factors drew us to Amazon EC2 DL1 instances based on Habana Gaudi AI accelerators,” said Ryan Ferguson, CEO of Riskfuel. “First, we want to make sure our banking and insurance clients can run Riskfuel models that take advantage of the newest hardware. We found migrating our models to DL1 instances to be simple and straightforward—really, it was just a matter of changing a few lines of code. Second, training costs are a big component of our spending, and the promise of up to 40% improvement in price performance offers potentially substantial benefit to our bottom line.”