AI era datacenters need AI-ready Ethernet

How the latest Ethernet switches will accelerate next-gen cloud infrastructures equipped to run AI workloads

Sponsored Post The rapid breakout of Artificial Intelligence is driving business opportunities across verticals – but there's one sector for which AI presents some formidable challenges, and that's the datacenter industry itself.

The data-intensive IT resources required by AI workloads make unprecedented demands on datacenter facilities' physical infrastructure and switching fabrics. It's that AI training models will grow in complexity by 1,000 times over the next three years – each model has billions of dense parameters, and might running for weeks on end.

The pressure of burgeoning AI workloads is revealing that many datacenter infrastructures are not up to the job. And without rapid rebuilds, many datacenters could be left having to limit the AI work they can take on.

Switching innovation is critical to re-architecting datacenters with accelerated infrastructure for the AI era. Typically, their interconnected servers are linked to the global digital network through a fabric of high-bandwidth switches. This allows datacenter operators to build highly scalable networks while also mixing and matching vendor products to create network architectures that are specific to their needs.

For this reason, the surging uptake of AI and cloud is driving foundational changes in Ethernet switching technology, pushing it to greater levels of performance.

Click to play

Marvell believes AI has fundamentally changed the network switching landscape. Its Teralynx 10 51.2 Tbps Ethernet switch has been designed to meet this emerging requirement for accelerated infrastructure. This advanced switching platform gives Marvell a lead in its market says the vendor, but also prompts some profound questions for the company and its customers. They include:

– Why do we need to re-architect cloud networks for AI workloads? Can't datacenters just continue to expand and beef-up the general-purpose cloud network to meet the extra demands of AI?

– Why does Marvell believe that the network operating system SONiC – Software for Open Networking in the Cloud – is set to play an influential role in AI and cloud switching?

– And looking ahead, what overall impacts will developments such as SONiC, the rise of AI, and faster tech upgrade cycles, have on the IT sector?

Answering these and other questions in this Hot Seat interview from The Next Platform is Nick Kucharewski, Senior VP & General Manager, Network Switching Business Unit at Marvell. Nick explains how, with its 51.2 Tbps Ethernet switch architecture, Marvell is plotting the industry roadmap for AI cloud switching.

You can view this free video by clicking the link above.

Sponsored by Marvell.