Intel-owned Cnvrg this week extended support for the chipmaker’s Habana Gaudi artificial intelligence (AI) accelerators.

The integration promises to dramatically reduce the cost of training deep learning and machine learning models by pairing Gaudi accelerators hosted in Amazon Web Services (AWS) with Cnvrg’s MetaCloud platform.

Guadi — Habana Lab’s second chip — is designed specifically with AI and machine learning training in mind. Intel, which acquired the company in late 2019 for $2 billion, claims the chips offer up to 40% better price to performance than completing GPU-based instances.

While Habana’s Guadi provides the AI muscle, Cnvrg’s MetaCloud platform is designed to make machine learning more accessible to data scientists by enabling them to build, manage, and automate deep-learning models across a combination of on-premises and cloud platforms running on CPUs, GPUs, and AI accelerators.

The Cnvrg-Habana Labs integration builds on an adjacent partnership with AWS, which saw the public cloud provider launch Gaudi-based instances at re:Invent 2020.

In fact, today’s announcement is predicated on AWS’s Elastic Compute Cloud DL1 Gaudi instance.

To establish a machine learning training environment, Cnvrg requires customers to first deploy a cluster in AWS Elastic Kubernetes Service (EKS) and then connect them back to MetaCloud. Once the integration is instantiated, customers can manage their cloud and on-premises workloads in one place.

Developers can then use Habana’s Synapse AI software development kit (SDK) to optimize their workloads for Gaudi. The SDK includes Habana’s graph compiler, runtime, tensor processor core kernel library, firmware, drivers, and development tools.

Habana Labs also offers examples of deep-learning training models that developers can download on GitHub to help jumpstart their deployments.

While the integration may streamline development on Habana's Gaudi chips, “it’s not that significant,” Karl Freund, principal analyst at Cambrian AI Research, told SDxCentral.

“Since Intel owns both companies, it’s really a chip shot,” he said. “The AWS integration is far more impactful.”

The integration would have been more significant, Freund argues, had Intel used Cnvrg to offer low- or no-cost access to its Gaudi AI accelerators to spur application development.

And enabling developers will be instrumental to driving Gaudi adoption, he added. “It takes time — too much time — to go from new silicon to real solutions. Gaudi is moving slowly because clients move slowly.”