Nvidia Corp. today announced broad availability of the next-generation of its DGX supercomputing platform, Nvidia DGX Cloud, giving companies access to thousands of graphics processing units on Oracle Cloud Infrastructure and its own cloud-based servers.

Announced at Nvidia’s GTX conference in March, DGX Cloud is a cloud-hosted artificial intelligence supercomputing service that provides immediate access to everything they need to start training the most advanced models for generative AI and other applications, Nvidia said.

The service is based on Nvidia’s popular DGX platform, which is a dedicated hardware offering companies can buy and set up to run in their own on-premises data centers. With DGX Cloud, enterprises no longer have to wait to purchase the expensive and often in-demand platform outright, but can instead rent the infrastructure they need for their AI workloads on a monthly basis. In other words, Nvidia is making its AI supercomputer available to many more enterprises, including those that cannot afford to deploy and manage its systems by themselves.

Nvidia explained that each instance of DGX Cloud provides access to eight of its 80-gigabyte Tensor Core GPUs, meaning 640 gigabytes of GPU memory per node. The platform is built atop a high-performance, low-latency networking fabric to ensure workloads can scale across clusters of interconnected systems. In this way, multiple DGX Cloud instances can act as one enormous GPU to tackle the most demanding workloads.

The DGX Cloud platform is paired with Nvidia’s AI Enterprise software, which gives customers access to more than 100 AI frameworks and pre-trained models, so they can build, refine and operate customized large language and generative AI models trained on their own proprietary data, for unique, domain-specific tasks.

Nvidia also provides access to its Base Command software for managing and monitoring training workloads on DGX Cloud. This also ensures DGX Cloud can work in tandem with on-premises DGX platforms, allowing enterprises to combine these resources when the need arises.

Nvidia had previously announced that prices for a single DGX Cloud instance start at $36,999 per month, and can scale up instantly as the customer requires.

DGX Cloud is launching on Oracle Corp.’s cloud infrastructure first and will later be offered on Microsoft Corp.’s Azure platform, followed by Google Cloud. There’s no mention yet if cloud industry leader Amazon Web Services Inc. will also host DGX Cloud.

With enterprises wanting to build bigger and better generative AI models, single GPU clusters are no longer enough to satisfy them, so a supercomputer makes a lot of sense, said Andy Thurai, vice president and principal analyst at Constellation Research Inc. He pointed out that Nvidia is not alone in going down the supercomputer route, as companies such as Hewlett Packard Enterprise Co., with its Cray computers, and startups such as Cerebras Systems Inc. have demonstrated.

Thurai said Nvidia, despite its dominant position in AI training, may actually face some tough competition from these rivals. “In HPE’s case, its GreenLake for Large Language Models offering may be more appealing as it is more eco-friendly with a near-zero carbon footprint,” he said. “Comparatively, Nvidia’s supercomputer clusters are a massive energy drain.”

That may make the difference in certain high-performance computing AI use cases, such as genome analysis, drug discovery and protein models, which all require high-intensity compute power, Thurai pointed out.

Still, Nvidia remains by far and away the most important provider of AI training infrastructure, and so DGX Cloud looks to be a strategic move as much as anything else. “Given the short supply and extremely high demand for GPUs, rather than selling chips and making trillions, Nvidia wants to own those chips and rent them as-a-service to make gazillions of dollars instead,” Thurai explained.

However, Thurai questioned whether Nvidia’s decision to rent DGX Cloud out on a monthly basis might be too much for some customers. In contrast, HPE and Cerebras are both planning to offer their services through a pay-as-you-go model. “Monthly rental might appeal to some very large language model builders, but it is a bloated and very expensive proposition that will seen as overkill by many enterprises,” he said.

In any case, Nvidia said it’s confident the availability of DGX Cloud will be a massive boon for generative AI, accelerating its potential use cases. It pointed out how early adopters of the platform have already recorded some impressive achievements.

For instance, healthcare firms have been using DGX Cloud to train protein models and accelerate drug discovery and clinical reporting. Meanwhile, financial services firms are using the platform to optimize portfolios, forecast trends, build recommendation engines and intelligent chatbots. Insurance companies are also using DGX Cloud to build models that can automate much of the claims process.

Analyst Patrick Moorhead of Moor Insights & Strategy said in remarks provided by Nvidia that rapid adoption of generative AI has become a key business imperative for enterprises, and many of them have been eagerly awaiting the launch of DGX Cloud. “The availability of Nvidia DGX Cloud provides a new pool of AI supercomputing resources with nearly instantaneous access,” he added.

Image: Nvidia