VMware’s management plane for distributed compute acceleration, Project Monterey, moved into early access this week.

Announced late last year, and in development for nearly three years, the project provides a consistent platform for distributed compute across multiple hardware accelerators using data processing units (DPUs), like those supplied by Intel and Nvidia.

According to Motti Beck, senior director of enterprise market development at Nvidia Mellanox, the growth of complex workloads like artificial intelligence (AI), real-time data streaming analysis, and security are putting an increased load on data centers.

“This reduces data center efficiency and can prevent IT from meeting its service-level agreements,” he wrote in a recent blog post.

DPUs promise to address these challenges by offloading network, storage, or security services from the CPU, freeing up additional compute resources.

But while DPUs offer numerous advantages, Baron Fung, research director at Dell’Oro Group, notes they are “not a drop-in solution to replace standard Ethernet adapters,” and require extensive software development to adapt them to the user’s infrastructure.

For the cloud providers and hyperscalers — Amazon and Microsoft, for example, have been deploying DPUs for years — this isn’t a problem thanks to their extensive engineering resources, he explained.

This is where Project Monterey comes in. The platform works as a management plane for orchestrating compute resources using DPUs, like Nvidia’s BlueField-2. In essence, this allows users to stitch together virtual machines (VMs) from resources distributed across the data center rather than being limited to a single server.

Traditionally smartNICs have been used almost exclusively for network monitoring and performance offload, explained Paul Turner, VP of product management for VMware’s Cloud Platform Business Unit, in an earlier interview.

“We think that’s really myopic on the capability there,” he said. “By putting ESXi managed through vCenter, we can start building a distributed virtual machine. To do that, we need to have a control software that actually can manage aggregating and pooling of resources.”

This capability is what makes VMware’s collaboration with Nvidia and Intel so significant, Fung said.

“The end result is that server utilization is further improved with a combination of reduced overhead and virtualization of resources with the VMware stack,” Fung explained. This benefits “both public cloud service providers and private enterprise data centers by streamlining the physical data center footprint.”

Over the past year, several vendors, including Marvell, Xilinx, Fungible, Intel, and Nvidia Mellanox, have brought new DPUs — sometimes called smartNICs or infrastructure processing units (IPU) — to market.

Nvidia’s third-generation BlueField DPUs, announced this spring, promise to offload up to 300 CPU cores worth of I/O tasks. The new chip is capable of running the entire VMware ESXi virtualization stack through Project Monterey, according to Justin Boitano, VP and GM of enterprise and edge computing at Nvidia

Late this spring Intel unveiled its next generation of infrastructure processing units — what the chipmaker is calling its DPUs — at the Six Five Summit.

Unlike Nvidia and Intel, composable infrastructure vendor Fungible has built its platform around a combination of server hardware, PCIe add-in cards, and its own software platform to achieve a similar goal to Project Monterey.

The company’s platform allows users to compose resources across physical servers using standard Ethernet using the transport control protocol.