At Dell Technologies World in Las Vegas, the emphasis was clearly on AI as the driving force behind the company's diverse offerings. CTO John Roese explained that AI's architectural needs significantly differ from traditional cloud infrastructure, prompting businesses to rethink their IT setups. He suggested that edge computing has shifted to primarily support AI applications, especially in various sectors like healthcare and transportation. Furthermore, while training AI models requires immense computational power, the inference process demands less in computation but involves high transaction rates with data systems, indicating a complex operational landscape for businesses.
[B]ecause we had a cloud infrastructure, we assumed that, well, maybe we'd use that for AI. And what we learned over the last two years is they are so different. They are so architecturally different, they have a such a different relationship with data, that the idea that five years ago, with no knowledge of AI, the decision you made as an enterprise to describe your IT infrastructure would be a perfect fit for a technology that emerged five years later that will transform the world, is insane.
You're going to need edge compute nodes and we've been doing edge for a long time, but in the last year and a half, I will tell you the dominant use of edge compute is AI now. There's very few people running other stuff on edge compute, you're running computer vision models, interpretation models. It's not necessarily generative AI, but you're doing AI there in factories and hospitals and transportation networks.
Training of AI models requires, actually, surprisingly less information than you'd think. But it requires the most incredible amount of compute you could imagine. Inference, on the other hand, funnily enough if you're doing RAG, is actually a lot lighter on compute but the transactions into the vector databases and the data systems are quite high. And so you end up with this kind of asymmetric behavior all over the place.
Collection
[
|
...
]