Nvidia’s DGX Cloud on OCI Now Available for Generative AI Training

Nvidia today announced broad accessibility for its cloud-based AI supercomputing service, DGX Cloud. The service gives users access to thousands of virtual Nvidia GPUs on Oracle Cloud Infrastructure (OCI), in addition to US and UK infrastructure.

DGX Cloud was announced at Nvidia’s GTC conference in March. It promised to provide companies with the infrastructure and software they need to train advanced models in generative AI and other areas that rely on AI.

Nvidia said this dedicated infrastructure is designed to meet Gen AI’s demand for massive AI supercomputing to train large and complex models like language models.

>>Follow VentureBeat’s Coverage of Continuously Generated AI<

“Like many businesses deploying DGX SuperPODs on-premises, DGX Cloud leverages the best computing architecture, with large clusters of dedicated DGX Cloud instances interconnected by the ultra-high bandwidth, low latency Nvidia network fabric,” Tony Paikeday, Senior Director of DGX Platform at Nvidia, told VentureBeat.

Paikeday said DGX Cloud simplifies managing complex infrastructure and provides a user-friendly “serverless AI” experience. This allows developers to focus more quickly on running experiments, building prototypes, and delivering workable models without the burden of infrastructure.

“Organizations that needed to develop generative AI models prior to the advent of DGX Cloud would have had on-premises data center infrastructure as their only viable option to handle such large-scale workloads,” Paikeday told VentureBeat. “DGX Cloud will enable any organization to remotely access its own AI supercomputer for training large and complex LLM and other generative AI models from the convenience of a browser without having to operate a supercomputing data center.”

>>Don’t miss our special issue: The Future of the Data Center: Meeting Growing Demands<

Nvidia claims the product enables generative AI developers to distribute massive workloads across multiple compute nodes in parallel, improving training speeds by two to three times compared to traditional cloud computing.

The company also claims that DGX Cloud will allow companies to establish their own “AI Center of Excellence” to support large teams of developers working on many AI projects simultaneously. These projects can benefit from a pool of supercomputing power that automatically accommodates AI workloads as needed.

Mitigate enterprise-generated AI workloads through DGX Cloud

According to McKinsey, generative AI could contribute over $4 trillion annually to the global economy by transforming unique business knowledge into next-generation AI applications.

The exponential growth of generative AI is forcing leading companies across industries to adopt AI as a business imperative, driving demand for accelerated computing infrastructure. Nvidia said it has optimized the DGX Cloud’s architecture to meet these growing compute demands.

Nvidia’s Paikeday said developers often face challenges in preparing data, building early prototypes, and using GPU infrastructure efficiently. His DGX Cloud, powered by the Nvidia Base Command Platform and his Nvidia AI Enterprise, aims to address these issues.

“Through the Nvidia Base Command Platform and Nvidia AI Enterprise, DGX Cloud helps developers get to production-ready models faster and with less effort thanks to accelerated data science libraries, an optimized AI framework, a suite of pre-trained AI models, and workflow management software that accelerates model creation,” Paikeday told VentureBeat.

Biotech company Amgen accelerates drug discovery with DGX Cloud. According to Nvidia, the company employs his DGX Cloud in combination with his Nvidia BioNeMo Large Language Model (LLM) software and his Nvidia AI Enterprise software, including his Nvidia RAPIDS data science acceleration library.

“With Nvidia DGX Cloud and Nvidia BioNeMo, our researchers no longer have to deal with AI infrastructure or set up ML engineering, and can focus on deeper biology,” said Peter Granthard, Executive Director of Biologics Therapeutic Discovery Research at Amgen’s Center for Research Acceleration through Digital Innovation, in a statement.

sound case studies

Amgen claims that the DGX Cloud has enabled the rapid analysis of trillions of antibody sequences, enabling rapid development of synthetic proteins. The company reported that DGX Cloud’s compute and multi-node capabilities resulted in 3x faster training of protein LLM with BioNeMo and up to 100x faster post-training analysis with Nvidia RAPIDS compared to alternative platforms.

Nvidia offers DGX Cloud instances on a monthly rental basis. Each instance is powered by 8 powerful Nvidia 80GB Tensor Core GPUs, providing 640GB of GPU memory per node.

The system uses a high-performance, low-latency fabric that enables workload scaling across interconnected clusters, effectively turning multiple instances into a large integrated GPU. In addition, DGX Cloud comes with high-performance storage to provide a comprehensive solution.

The product also includes Nvidia AI Enterprise, a software layer with over 100 end-to-end AI frameworks and pre-trained models. The software is intended to help accelerate data science pipelines and accelerate production AI development and deployment.

“DGX Cloud not only provides massive compute resources, but it also helps data scientists be more productive and utilize resources more efficiently,” said Paikeday. “With the help of his AI experts at Nvidia who help optimize customer code and workloads, you can get started quickly, launch multiple jobs simultaneously with great visibility, and run multiple generative AI programs in parallel.”