Virtualization software giant VMware Inc. said today at its VMworld 2020 virtual event that it’s partnering with computer graphics chipmaker Nvidia Corp. on a new enterprise platform for artificial intelligence workloads.

VMware also announced a new architecture for the cloud, data center and edge that runs on Nvidia’s newest data center processing units.

Nvidia will integrate its Nvidia GPU Cloud or NGC AI processing software with VMware’s main virtualization platform vSphere. NGC will also integrate with VMware Cloud Foundation, a hybrid cloud platform for managing virtual machines and software containers, and VMware Tanzu, a platform for building, running and managing Kubernetes-controlled container applications.

“We are partnering with Nvidia to bring AI to every enterprise; a true democratization of one of
the most powerful technologies,” said VMware Chief Executive Pat Gelsinger (pictured, right, with Nvidia CEO Jensen Huang).

VMware said these integrations will create an “enterprise-ready AI platform” that will enable companies to run new AI workloads and manage them via a single platform. It will also give companies a way to deploy AI-ready infrastructure where their data resides, be it in the cloud, in the data center or at the edge of the network.

The Nvidia GPU Cloud is a comprehensive catalog of optimized software tools for deep learning and high-performance computing that take full advantage of Nvidia graphics processing units.

Integrating this software with VMware’s tools will make it much easier for enterprises to deploy new AI workloads on their existing infrastructure, the companies said. It will give data scientists and developers access to a wide range of cloud-native GPU-optimized containers, AI models and industry-specific software development kits.

That’s important because Nvidia is widely considered to be one of the industry leaders in AI, not only with its hardware, but also thanks to its software.

“As much as people think of us as a hardware company, we are actually a software company too,” Manuvir Das, Nvidia’s head of Enterprise Computing, said in a press briefing. “Over 2.3 million developers work with the Nvidia ecosystem.”

Das said that it will now be possible for enterprises to create cloud-native AI apps made using NGC software on VMware vSphere and Cloud Foundation with the Tanzu platform. Meanwhile developers will benefit from having a common experience, no matter if they’re using containers or virtual machines to host those applications.

“The beautiful thing about vSphere is that you get the same experience whether you run on VMs or Kubernetes,” Das said. “Even without a container platform, you can host your app in a container for the convenience of packaging. We anticipate a good percentage of customers deploying in that fashion.”

The companies also announced a new initiative called Project Monterey that will see them work together to build a hybrid cloud architecture. It’s based on Nvidia’s smart network interface controller technology, called SmartNIC, which incorporates its newest programmable BlueField-2 data processing units.

The idea is to combine the BlueField-2 DPUs with VMware Cloud Foundation to create a next-generation architecture that’s better able to handle the unique demands of AI, machine learning, high-throughput and data-centric apps, the companies said. The SmartNIC controller helps to accelerate application workloads and provide an extra layer of security by offloading hypervisor, networking, security and storage tasks from central processing units to DPUs.

The collaboration recognizes that Nvidia’s SmartNIC platform and its DPUs are critical technologies for accelerating applications where the data resides, the companies said.

“Project Monterey reimagines data center infrastructure for hybrid cloud apps,” said Krish Prasad, senior vice president and general manager of Cloud Platform Business at VMware. “Cloud Foundation users will be able to speed up a wide range of next-generation and general-purpose applications, deliver programmable intelligence and operate a distributed zero-trust security model across data centers, the edge and telco clouds.”

The biggest appeal for enterprises is that Nvidia’s architecture should enable big application performance gains. Das said Nvidia has repeated much of VMware’s work to minimize virtualization overheads with its GPUs. That should help to expose VMware workloads to the raw capabilities of the GPU and lead to dramatic performance improvements.

“For example, using AI for breast cancer studies, when you run it on the same environment with and without GPUs, you get a performance gain of 31 times,” Das said.

Customers will also benefit from a much more flexible infrastructure, Das said. Customers will be able to take advantage of an Nvidia technology called multi-instance GPUs that enables them to take a GPU and slice it into smaller parts, allowing them to mix and match that infrastructure for tasks such as data preparation, training or inference.

Analyst Holger Mueller of Constellation Research Inc., who was briefed on the new partnership, told SiliconANGLE that enterprises today are eager to move faster and become more agile, and that Nvidia and VMware are playing to this need. He said these kinds of partnerships pave the way towards enterprise acceleration, as they provide pre-configured out of the box capabilities for companies that don’t require any integration work.

“The new offerings are a win-win-win for everyone concerned,” Mueller said. “VMware wins because its customers get access to Nvidia’s popular GPUs and DPUs, Nvidia wins because it gets access to VMware’s ecosystem with a productized offering, and most importantly enterprises win because they get a more powerful platform for their next-generation applications.

With reporting from Paul Gillin

Photo: Nvidia/VMware