VMware Adds Nvidia AI Software Support to vSphere

The early March 2021 release of vSphere 7 Update 2 saw VMware making a big step into the world of artificial intelligence and machine learning: support for Nvidia AI software has now come to its flagship platform.

This wasn't entirely unexpected. During September's online VMworld event, Nvidia and VMware announced a partnership, with a goal of making Nvidia's AI Enterprise software available to VMware's platform for deploying and managing virtual machines.

With this latest release, vSphere is "exclusively certified" to run Nvidia's AI Enterprise's applications and frameworks, which have now been containerized so they can be run across an organization's infrastructure instead of being held in a silo. This naturally also means increased support for Nvidia GPUs, which are required to run the software.

"It opens up for both of us a good opportunity," Lee Caswell, vice president of marketing at VMware, told ITPro Today. "We're looking to go and help AI become mainstream in the enterprise. They're looking to open that up for all of our 300,000 vSphere customers, who can now have access to these new capabilities."

Having access to this market is important to Nvidia because it helps them bolster sales in their datacenter division, which pulled in $6.7 billion in the past fiscal year.

Getting a toehold in the emerging enterprise AI market is also important to VMware, which has spent the last several years expanding its offerings beyond the virtualization technology it pioneered, mainly through acquisitions; in 2018, VMware added cloud-native technology to its portfolio with purchase of the Kubernetes startup Heptio and about a year later bought back Pivotal, a cloud native platform provider it had spun-off in 2013.

Nvidia and Vsphere

Typically AI software has been run on bare metal, Caswell said, in order to avoid potential performance degradation associated with bringing compute-heavy workloads to VMs or containers. The problem with this approach is that bare metal deployments are not portable. As a result, AI workloads are confined to silos, which is problematic for enterprises wishing to harness AI on-the-fly across their IT infrastructure.

By taking advantage of properties inherent in vSphere's hypervisor, VMware and Nvidia have been able to containerize Nvidia's AI Enterprise software with nearly the same benchmarked performance levels as running on bare metal. This makes Nvidia AI software readily accessible throughout an organization’s infrastructure, this solving the problem of maintaining portability without suffering significant performance degradation.

In order for the AI software to run properly, it needs to be able to take advantage of GPUs, which takes much of the load off of a servers' CPUs by doing most of the heavy lifting. For this, VMware has added support for the Nvidia's A100 Tensor Core GPUs that are included in Nvidia-Certified Systems, Nvidia tested and approved server designs that are currently being marketed by eight equipment makers, including ASUS, Dell EMC, HPE and Supermicro.

In addition to running AI workloads, GPUs can also be used by other vSphere features, such as Multi-Instance GPU, which allows GPU cycles to be shared by multiple users, and Distributed Resource Scheduler for automatic workload placement to avoid performance bottlenecks.

"Up to seven VMs can now share a single GPU," Caswell said. "That's a more cost-effective way to deploy at the enterprise."

Load Balancing Containers and More

There are more improvements to vSphere than support for Nvidia AI software, many of them revolving around using vSphere with Tanzu, the latter being VMware's Kubernetes offering, which is  available as an addon for vSphere. According to VMware, this latest release running Tanzu on vSphere offers a faster, more scalable and secure application experience because of the addition of NSX Advanced Load Balancer Essentials.

Caswell said that vSphere had previously relied on HAProxy, an open source load balancer, which has now been replaced by the NSX Advanced Load Balancer that VMware acquired when it bought Avi Networks in 2019 and which has now been integrated into vSphere.

"It gives you more control," Caswell said. "No charge for this, but you're getting more integrated management over time, extended down into the load balancing capabilities."

VMware has also made many integration improvements for those using vSphere with the company's vSAN storage virtualization software. For example, there's the addition of enhanced HCI Mesh, which makes it easier to share storage from multiple servers, and for security, data can now be encrypted in transit between compute and storage devices.