With effectiveness, reliability and trust remaining critical challenges in artificial intelligence, validated designs in AI ensure transparency and accountability, simplifying compliance with regulations and ethical standards.

Validated designs accelerate innovation by transforming AI from a risky experiment into a reliable tool. They ensure efficiency and trust, making AI ready for confident deployment, according to Zeus Kerravala (pictured), founder and principal analyst of ZK Research Inc.

ZK Research’s Zeus Kerravala talks with theCUBE about validated designs in AI.

“With AI you will see companies, you see Cisco and Nvidia working together, Pure Storage and Arista having a partnership with Nvidia as well and build another turnkey system,” Kerravala said. “These turnkey systems to me are the right way to go for an enterprise because when you start to deploy these things, there’s a lot of dials you can turn and levers you can pull in order to tweak the systems. That’s why the validated designs are important.”

Kerravala spoke with theCUBE’s Dave Vellante at theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how validated designs in AI drive innovation and accelerate adoption.

What it takes for validated designs in AI to be successful

Validated designs thrive on the synergy of fast storage, high-speed networks and powerful processors. Fast storage ensures instant access to massive datasets, high-speed networks enable seamless system communication and advanced processors provide the computational power for complex model training and real-time inference, Kerravala explained.

“You need three things to make sure your AI works, fast storage, fast network and fast processor,” he said. “Everybody understands the role of GPUs. If your network speeds don’t match what you’re trying to do to the GPU layer and your data access speeds on the storage side don’t match that, one of those as a weak link will create the whole system to not work properly. The three of those things have to be in lockstep with one another.”

InfiniBand delivers ultra-low latency and high-speed connectivity ideal for tightly coupled AI workloads, such as distributed deep learning, where every millisecond counts. Ethernet offers cost-effective, flexible scaling for AI inference, data ingestion and storage networking. Together, they create AI infrastructure that is both high-performance and adaptable, supporting larger models and growing datasets, Kerravala pointed out.

“If you roll back the clock 30 years, there were all these different protocols and Ethernet beat them all,” he said. “InfiniBand is the last band standing. You can have InfiniBand grow and Ethernet grow at the same time because we’re seeing AI scale-up, scale-out, scale-across. I think over time what happens is InfiniBand gets used for cases where it’s a fully integrated system, something like NVLink where I’m connecting GPUs within a box. Ethernet’s a better system for that because it’s something people are familiar with and ought to work with.”

The rise of neocloud providers prioritizing regional data residency marks a major shift in cloud computing, driven by surging AI workloads. By creating a specialized layer for AI computation, neoclouds introduce a new tier in cloud architecture — one poised to evolve alongside advancing AI, according to Kerravala.

“I think initially it’s the cloud, neocloud, but I do think enterprises need to start thinking about it too,” he said. “If you’re a Global 2000, I do think that you do need to think about how to re-architect your data centers. For a while everybody just worried about scale-out and that’s why the leaf spine architecture became so popular. I think within these clusters now, you do have to worry about vertically integrated performance as well as a performance across racks.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event: 

Photo: SiliconANGLE