Nvidia’s DGX Cloud brings AI supercomputing to enterprises via a web browser

Nvidia announced today at its annual GPU Technology Conference (GTC) that it’s now offering a service that will allow enterprises to access its AI supercomputing infrastructure and software through a web browser.

The service — called DGX Cloud — provides dedicated clusters of Nvidia’s DGX AI supercomputers, which are designed for deep learning and other advanced AI applications. The company said that DGX Cloud would simplify the complexity of acquiring, deploying and managing on-premises infrastructure for AI.

>>Follow VentureBeat’s ongoing Nvidia GTC spring 2023 coverage<<

Nvidia has been iterating on different DGX form factors over the last several years, including building out small units with the DGX Station in 2021. The DGX Cloud now takes Nvidia hardware to the cloud, where it is offered as a service to offer organizations the full power of the converged platform, putting it in the cloud and making it accessible via a web browser interface.

“What we’ve done over the years with DGX is not just build the state-of-the-art supercomputer, but we built a software stack that sits on top of it, that turns this into a turnkey training-as-a-service platform,” Manuvir Das, VP of enterprise computing at Nvidia, said during a press briefing. “You just provide your job, point to your dataset, hit go and all of the orchestration and everything underneath is taken care of.”

Early users of DGX Cloud claim improved training speed

While the public service is only being formally announced now, Nvidia has had some early users of the DGX Cloud.

Among them is biopharmaceutical vendor Amgen, which has been using the DGX Cloud to accelerate machine learning (ML) training speed. Das noted that Amgen has reported three times faster performance for its training using DGX cloud versus other approaches.

Insurance technology vendor CCC Intelligent Solutions (CCC) is also a DGX Cloud user. Das said that CCC is an insurance company that is using AI to process claims from accidents in a matter of seconds versus days. CCC is using DGX Cloud to help with scaling demand.

“We run some experiments on-premises on Nvidia DGX systems, but we may have spikes where we want to add, for example, 10 million more data points and do another run,” Neda Hantehzadeh, director of data science at CCC, wrote in a blog post. “If we need additional capacity, we can switch to DGX Cloud.”

DGX Cloud service lands in the public cloud with Bluefield 3 DPU

DGX Cloud is being delivered as a service in the public cloud.

Das said that it is currently available on Oracle Cloud Infrastructure (OCI) as well as Equinix. He added that it’s coming soon to Microsoft Azure and Google Cloud.

The Oracle deployment was highlighted in a press briefing by Kevin Deierling, VP of networking at Nvidia. As part of its DGX Cloud deployment, Oracle is also making use of Nvidia’s Bluefield-3 data processing unit (DPU). Nvidia first announced the Bluefield-3 back in April 2021, but it has taken a couple of years to bring to full production. A DPU is a dedicated computing unit for data processing, which helps to accelerate overall performance by reducing the overhead on CPU and GPU resources.

“The Bluefield-3 has twice as many ARM processor cores, more accelerators, and runs workloads up to eight times faster than our previous generation DPU,” Deierling said. “The Bluefield-3 offloads, accelerates and isolates workloads across cloud, HPC, enterprise and accelerated AI use cases.”