OPI Project aims to standardize DPUs and IPUs for industry adoption

In recent years, silicon vendors have been building out new types of computing architecture beyond just CPUs and GPUs – welcome to the world of data processing units (DPUs) and infrastructure processing units (IPUs).

The goal with DPUs and IPUs is to let organizations offload certain data and cryptography as well as artificial intelligence/machine language (AI/ML) tasks to dedicated hardware to accelerate operations. To date, there have been few, if any, standards around DPUs and IPUs to enable interoperability or industry standardization for deployment, management and scheduling, but that’s about to change.

Today, the Linux Foundation announced the launch of the Open Programmable Infrastructure Project, which aims to collect open-source efforts around DPUs and IPUs and organize vendors to advance adoption for organizations of all sizes. Founding members of the Open Programmable Infrastructure (OPI) project include Intel, Nvidia, Marvell, F5, Red Hat, Dell and Keysight Technologies.

Among the initial projects set to become part of the OPI is the IPDK (infrastructure programmer development kit) that is being developed by Intel and the Diamond Bluff project that is being built by Red Hat and F5.

“Red Hat and F5 and a few other companies were working on the Diamond Bluff project and we [Intel] started talking to them and we soon realized that we have similar goals and we should work together,” Kyle Mestery, senior principal engineer at Intel, told VentureBeat. “The goal of Open Programmable Infrastructure Project is to foster a community of open standards and open-source projects around these next-generation architectures, which include DPUs and IPUs.” 

Market for DPUs and IPUs is new but growing

Both IPUs and DPUs have found their way into the architectures and deployment of hyperscalers and cloud providers in recent years.

The opportunity for the hyperscalers and cloud providers has been to provide a more granular level of services for users to consume. There is also potential for IPUs and DPUs to help enterprise users as well, which is one of the goals that Mestery said he hopes the OPI will achieve in the coming months and years.

Intel has been building out infrastructure processing units (IPUs) in recent years as a form of silicon hardware technology that makes use of field programmable gate arrays (FPGA) and application-specific integrated circuits (ASIC) IPU technology from Intel.

Nvidia has also been active in the space building out its BlueField DPUs that are focused on accelerating analytics and AI workloads. Marvell is also active in the market with its Octeon DPU technology, which can be used to help to support 5G data workloads.

How to manage DPUs and IPUs at scale

Both the IPDK and the Diamond Bluff projects involve looking at different aspects of how an organization can provision IPUs and DPUs in a data center deployment.

The OPI project is also talking about the application programming interface (API) layer for IPU and DPU management to help organizations have a unified approach to visibility and control. Mestery emphasized, though, that the OPI isn’t looking to reinvent technologies and will be collaborating with other Linux Foundation initiatives including the LF Networking project and the Cloud Native Computing Foundation (CNCF).

Mestery cited the OpenTelemetry project as an example from the CNCF. Its goal is to collect observability data on running operations. He noted that if an OPI open-source project needs  telemetry data, it makes sense to leverage OpenTelemetry rather than create something new.

As the OPI ramps up, Mestery said he is hopeful that the open-source effort will be able to demonstrate the benefits of what DPU and IPU technology can provide. As a goal, he wants the project to make it easier for organizations to take DPU and IPU technologies and deploy them inside of their own private data centers and use them for a number of purposes, whether it’s for edge computing, private cloud or whatever they want to do with it.

“I hope that we can provide a framework that makes these technologies easier for organizations to use,” Mestery said.