How Nvidia Is Helping Partners ‘Democratize AI’ For Enterprises

The chipmaker is making a major push for GPU-accelerated computing in enterprises this year, and it’s taking notes from VMware to simplify the experience for partners and customers. Nvidia’s head of enterprise computing, Manuvir Das, talks to CRN about how the company’s new software and hardware enterprise solutions will create new opportunities for channel partners.

Nvidia believes the stars have aligned for enterprise adoption of AI thanks to a combination of certified GPU servers, a new kind of computer chip that can improve data center economics and IT-friendly AI software that has been modeled after VMware.

Manuvir Das, head of enterprise computing at Nvidia, told CRN that with these elements, the chipmaker this year is making its biggest sales push yet with channel partners and OEMs into enterprise customers. The goal, he said, is to make AI and data analytics applications and other kinds of GPU-accelerated software as accessible and manageable as the rest of the data center.

[Related: Nvidia Rival Graphcore Makes Big North American Partner Push: Exclusive]

“The big shift is, for the first time, we’re saying that for enterprise customers, we’re ready to democratize AI, and we’re ready to make this thing usable by every enterprise customer,” said Das, who was previously an executive at Dell EMC for seven years.

Crucial to this push with partners is the Nvidia AI Enterprise software suite, which allows customers to run GPU-accelerated applications in existing data center infrastructure with VMware vSphere. Equally important are the new Nvidia-certified servers from OEMs that are guaranteed to run Nvidia AI Enterprise’s applications, tools and frameworks at optimal levels.

“The way we are propagating this with the channel and the OEMs is as a solution bundle, where it’s the certified server and vSphere license and Nvidia AI Enterprise license,” said Das, who added that the Nvidia AI Enterprise license pricing is modeled after vSphere.

Launched as part of an expanded alliance with VMware earlier this year, Nvidia AI Enterprise promises to offer near bare-metal performance for GPU-accelerated applications while allowing users to spin up GPU compute resources into containers and virtual machines on vSphere.

Das said he thinks Nvidia AI Enterprise will present new services revenue opportunities for partners, particularly in the area of ML Ops, short for machine learning model operationalization management. He believes this because Nvidia AI Enterprise takes care of most of the behind-the-scenes plumbing that’s necessary to manage and track changes for machine learning models.

“We’ve done all that work, so ecosystem, please arrive, do interesting solutions for ML Ops incrementally on top of what we got and go sell that and drive revenue from that,” Das said.

He likened the opportunities presented by Nvidia AI Enterprise to those that have been created by Microsoft’s Windows Server operating system.

“It’s not that different from something like Windows Server, where the model is Windows provides a whole bunch of capabilities and there’s been a huge ecosystem of [independent software vendors] that have put software on top of that,” Das said.

For those reasons, Das said, Nvidia is making it a “huge priority” for his company’s channel partners to sell Nvidia AI Enterprise.

“We are now at the point, where we feel the conditions are right, not just because there’s maturity in the software and the hardware, but also, we work with customers a lot, so from a customer point of view, we really see this now,” he said.

Das said Nvidia has seen growing demand for GPUs in cloud instances, which, to him, is one major signal that there is an appetite among enterprises. At the same time, he added, there are a significant number of customers in verticals like financial services and health care who don’t want to put their data in the cloud and therefore need on-premises solutions for GPU-accelerated computing.

But Nvidia recognizes that not every enterprise needs an expensive server with four or eight GPUs, which is why the company is pushing its EGX platform as a more affordable option. Atos, Dell Technologies, Gigabyte, Lenovo and Supermicro all announced their own EGX servers at Nvidia’s GTC 2021 event earlier this year, allowing enterprises “to run AI workloads on the same infrastructure used for traditional business applications,” according to Nvidia.

“What we’ve done here with EGX is we worked with the OEMs to incorporate GPUs into their volume servers, their 1U, their 2U [configurations],” Das said. “The Dell PowerEdge series, for example, is one. This is a server that a customer would procure for $10,000 to $12,000; it now has one GPU in it, and its cost goes up by like a couple of thousand dollars.”

While most of Nvidia’s focus in the hardware space for enterprises right now is GPU computing, the company is starting to ramp up production, sales and marketing for a new kind of computer chip, a data processing unit. Nvidia’s DPUs, which bear the brand BlueField, act as a SmartNIC, replacing the standard network interface card in servers, and can offload networking, storage and security workloads from the CPU while enabling new security and hypervisor capabilities.

The CPU offload capabilities can have major implications for a data center’s total cost of ownership, with the latest BlueFIeld-2 capable of offloading 30 CPU cores worth of workloads and next year’s BlueField-3 capable of offloading up to 300.

When enterprise customers need to refresh their servers, Das expects the TCO benefits of Nvidia’s DPUs will drive them to adopt the new kind of computer chip, and for that reason he thinks the BlueField-3, due out in early 2022, will be Nvidia’s first DPU to see mass adoption.

“I think in that case, the entire offloading capability will be what drives the choice of DPU in the refresh, because my servers can do 30 percent more work if I put a pretty cost-effective DPU in there, so, of course, that would be attractive,” he said.

But outside of a refresh, Das believes the BlueField’s security features, which includes real-time network visibility, detection and response capabilities, will be what drives customers to adopt DPUs.

“With the DPU, all the data is flowing through there anyway because the DPU, for starters, is a network interface card. It’s actually the NIC on the server, where all the packets are flowing through, so it’s very natural and efficient to inspect the packet right there while you’re already processing it,” he said.

VMware has already enabled Nvidia’s BlueField-2 DPUs to run the virtualization giant’s ESXi hypervisor, which Das said will create an “open platform” for software developers to develop new firewall and packet inspection applications, among other things, that can run on DPUs. That, in turn, will create new opportunities for channel partners.

“There’s a variety of use cases. For storage acceleration, there’s a lot you can do, not on the storage server, but on every application server that is talking to the storage,” Das said. “You can do client-side work on the DPU to accelerate the connection to the storage, so that’s another great application.”

Nvidia’s confidence in enterprise adoption of AI is reflected in the company’s earnings for the fourth quarter of its 2021 fiscal year, which ended Jan. 31 and was reported in late February. During the earnings call, Nvidia CEO Jensen Huang said data center GPU sales through OEMs to vertical industries grew faster than revenue from hyperscalers like Facebook and Google for the first in Nvidia’s history.

At the time, Huang called this next wave of AI adoption the “industrialization of AI” and said it represented the industry’s “smartphone moment.”

“All of these industries, whether you’re in medical imaging or in lawn mowers, you’re going to have data centers that are hosting your products, just like the [cloud service providers], and so that’s a brand-new industry,” he said in February.

Matthew DuBell, a consulting solutions architect for the Business and Analytics Advisors group at St. Louis, Mo.-based World Wide Technology, a top Nvidia partner, told CRN that the chipmaker has a “very holistic strategy” for the data center because of how Nvidia is bringing GPU-based solutions for AI and analytics to production and enabling new offload capabilities with DPUs. Those things will go a long way in making it simpler for enterprise customers to adopt AI hardware.

“I really think a lot of those tools will be easier to consume and for an organization to get started and see value return on investments inside of AI programs via AI Ops or IT or business-related use cases such as natural language processing, digital twins and simulation,” he said.

With the Nvidia-Certified Systems program and EGX platform, the chipmaker is removing any concerns enterprise customers might have about the performance of GPU-accelerated servers, DuBell said.

“Nvidia does very rigorous disciplined testing around these things, and they understand based on the price point that you’re ordering, here’s the expected performance,” he said. “So to me, when I look at an IT organization managing multiple servers, it’s critical that they don’t have to worry about this individual server’s performance meeting par.”

However, DuBell thinks it’s still early innings for Nvidia’s enterprise efforts.

“Nvidia is now looking at the data center as the whole compute model, and I think enterprise architects are starting to wrap their heads around what does that mean,” he said.

To prepare for AI adoption, enterprise customers will need to have a good grasp on their data operations and processes to ensure they can take advantage of the new capabilities, DuBell said. Once they do, the Nvidia-Certified Systems program and Nvidia AI Enterprise will play a major role.

“What we’re really seeing is people talking about it now, which is good, and I’m hopeful that within a few months [or by] the end of the year, we’ll start to see adoption and the technology coming into production,” he said.

As for DPUs, DuBell said he is seeing a lot of interest among customers who want to minimize their compute footprint in edge data centers. He likened the DPU’s offloading capabilities to how data center operators have focused on right-sizing virtual machines for several years now.

“If I can move the management of that into these specialized environments that are secure, that makes me even more efficient with my general compute clusters,” he said. “So I think that’s where a lot of organizations are wanting to see how it goes at the edge with the current DPUs, how pliable they are, how easy they are and then [they’ll] look at the advantages of BlueField-3.”