Nvidia launches integrated AI platform for VMware vSphere 7

Nvidia rolled out a suite of AI tools and frameworks certified to run exclusively on VMware vSphere 7 Update 2, allowing users of the virtualization platform to exploit capabilities of Nvidia's A100 Tensor Core GPUs.

The purpose of the Nvidia AI Enterprise suite is to help larger enterprises modernize their infrastructure to better meet the performance demands required by AI-based applications, according to both companies.

Through this partnership, first disclosed in September, the companies expect the new offering to enable IT pros to fast-track such applications into virtualized data centers, resulting in a more expansive AI-aware infrastructure.

The Nvidia AI Enterprise suite enables corporate users to develop a range of AI-based products and services, such as advanced diagnostics for the healthcare industry and fraud detection for financial services companies.

"To this point, AI has largely been an island where people had to try a DIY approach to set up and manage the technology," said Justin Boitano, vice president and general manager of enterprise and edge computing at Nvidia. "But with VMware's help, we can put together an infrastructure people can use with the range of VMware tools that already exist and are optimized for AI. We want this to be a turnkey system for admins to more easily support line-of-business teams."

The offering for vSphere 7 Update 2 allows users to support new AI implementations across large-scale data centers and hybrid clouds. The AI suite also provides scale-out, multinode AI application performance on vSphere 7 that is on par with bare-metal servers, according to the company.

"The world has typically run AI on bare-metal servers," Boitano said. Nvidia's AI Enterprise suite allows users to reduce AI model development time "from 80 weeks to eight weeks," and deploy the more advanced AI applications on vSphere with the same performance seen on bare metal, he said.

Bringing Nvidia's and VMware's core offerings closer together is a logical pairing that figures to benefit both companies, as well as their respective users looking to speed code-heavy AI applications, according to analysts.

A lot of AI technology relies on GPUs, so Nvidia providing the GPUs and vSphere is where the AI workloads will run; that can serve as a pretty complete integrated platform.

"This is really all about the integration of their products, having a sort of certified tested platform," said Gary Chen, analyst at IDC. "A lot of AI technology relies on GPUs, so Nvidia providing the GPUs and vSphere is where the AI workloads will run; that can serve as a pretty complete integrated platform."

Nvidia officials noted it has certified vSphere as the only virtualization software that will offer hypervisor support for live migration with Nvidia's Multi-Instance GPU technology. This allows each A100 GPU to be partitioned into as many as seven instances at the hardware level, they said.

"[VMware] users can now incorporate the latest GPUs into their virtual environments to take advantage of features like multi-instance GPUs, allowing GPU cycles to be shared across multiple VMware products," said Geoff Woollacott, senior strategy consultant and principal analyst at Technology Business Research.

In addition, Nvidia's ConnectX-6 adapters are certified for VMware vSAN over Remote Direct Memory Access (RDMA), which offloads CPU communication tasks to boost application performance.

The Nvidia AI Enterprise suite is available as a perpetual license at $3,595 per CPU socket, with Enterprise Business Standard Support for Nvidia AI Enterprise priced at $899 annually per license, per year. Users have the option to apply for early access to Nvidia AI Enterprise as they plan upgrades to VMware vSphere 7 Update 2.