Mark Papermaster on Advancing AI, Open Source, & Developer Impact

Well Bryan, today was such a fun day for us at AMD because we not only announced advancements in our portfolio, our new CPU for the data center, the 5th Generation EPYC. We announced our next advancement in our AI roadmap with the MI325. But moreover, we announced enhancements to our networking that put this all together, and the progress that we've made with partners. It's actually just incredible when you think about our announcement of Instinct, our AI Data Center GPU product, in December of 2023. Here we are in October, and we are just in production workloads across a broad set of applications. That's exciting. It's a proud day for us today at AMD.

Our focus has been, and what we've been doing throughout the transformation at AMD, it has been about bringing competition back to the industry and doing it in such a way that we're creating an open ecosystem. And so there are several parts of that. They come together and they create incredible momentum for us with our partners. The facets are, before you even get to the technology, we have a culture that's open and collaborative. So that's actually foundational for us. So the way that we partner with others is very deep. We listen. We take action when we get feedback from our customers. Actually, that came out at our Advancing AI event today from our partners on stage with us. So that's foundational, the way that we are open and we collaborate. And to how do you interconnect these high performance standards? How do you create a competition to InfiniBand with the Ultra Ethernet consortium that brings all the capabilities of high performance and the kind of congestion management you need for that traffic for this high performance AI? And we're doing it open. Open standards. And we showed our first network product that will be the first adopter of that open standard. So there's a whole hardware element that allows us to really partner and give customers choice. They're not locked in to any one solution. And they can mix and match so that, as Lisa always says, there's no one size fits all with these high performance computing, these high performance AI applications. And so, again, it's our heritage at AMD to open source. Our ROCm stack for AI has been open from its inception. And so we are partnering with the community, and it's incredibly powerful, and we get tremendous feedback. And so when you think about how is it that we announced this product MI300X for AI in December, and again, we have all of these applications running, absolute real business dependent workloads that are in production now, we couldn't have done it without the community.

I don't think people realize the critical role that developers play for AI to reach its promise. And why do I say that? Well, you think about where the progression has been today. It's been on these frontier models, these massive models, and it's been the type of ChatGPT, Gemini and the rest where these large language models, and then that spoken interface allows us all to get access to high performance computing, high performance AI. But what happens when AI needs to be built into almost everything? That's the generation of agentic AI where you're putting different types of models together, and you're having to tailor them to help productivity of a given task. So all of us, I think, realize we're at an inflection point where AI to serve real, productive, enhancing purpose is right upon us. And the reality will be delivered by creating a vast ecosystem of developers. And again, that's where we feel with our open-source platform, we're there. And you look at what we announced today, our partnership with Clem at Hugging Face, over a million open AI models available to developers. We're a founding member of PyTorch, the most widely used development platform. It's fully open.