As IBM began shifting its focus to scalable cloud-native architecture in 2019, it sought out infrastructure that could scale to the cloud's needs by providing high levels of performance, security and architecture resiliency. The vendor found its answer in AMD Pensando's data processing unit (DPU).

"We evaluated several smart NICs in the market at the time, and Pensando really showed up in a big way," IBM GM of Cloud Infrastructure Services Satinder Sethi told SDxCentral. "It met a lot of our requirements from a performance, security, compliance and [total cost of ownership] perspective."

According to AMD Corporate VP of Networking Technology and Solutions Group Soni Jiandani, Pensando allows customers to "democratiz[e] DPU technology with software and the ability to program it for all different workloads," she told SDxCentral. "It's really a future proof architecture," she argued.

The key to AMD Pensando's DPU technology is the programmable P4 language used in data and management planes. This gives AMD and its cloud customers, like IBM, the flexibility to accommodate the needs of traditional applications – and "new age" applications. "Otherwise, you get siloed," she said. "You get forced into building dedicated, unique things for new emerging applications."

AMD Pensando's programability was a major point of attraction for IBM. "As a cloud provider, we have our very unique requirements based on the kind of workloads that we are going after" – like highly regulated industries – and "the kind of security compliance posture" those workloads necessitate.

"Even if we started our journey with the best possible smart NIC out there, we knew that as we scaled and as we grew and brought on new customers, requirements would continue to evolve and change," Sethi explained.

This partnership has also enabled IBM to offer services that "we just couldn't do earlier," he said. These include congestion control in a scalable network, where cloud is traditionally implemented at spine and leaf architectures and at the edge. "Now we can actually implement that at a host level [by] leveraging the programmability of P4 with the engineering innovation that we're doing with Pensando," he explained. "Those elements are extremely critical as we look forward to scaling our environment with the assurance that we have a very, very scalable platform."

"It's not only about the technology," Jiandani said. "It's also about how we are innovating together with IBM."

The Pensando team's mission has always been to build a multi-generational DPU architecture that ensures "customers like IBM, who we are co-innovating with from a software stack perspective, have the ability to take that architecture to 400 gig and then 800 gig and beyond," Jiandani explained.

The secret to this partnership's success, she added, is founded on that multi-generational architecture and paired with feature rich software capabilities and an open, programmable architecture that facilitates co-innovation between the two companies. "We don't have a vendor-supplier relationship. We have a true partnership with IBM," Jiandani said.

That collaborative spirit can be felt all the way down to the partners' combined engineering teams. "You will not be able to tell [if] they are on IBM's payroll or our payroll. They work as a unified team, and they continue to innovate at breakneck speed," she said.