AMD on NVIDIA does not use GPU chiplets: ‘we have a big lead there’

AMD will debut its next-gen RDNA 3 architecture later this year, where the big boy flagship Navi 31-based offering is rumored to have a GPU chiplet-based design… and NVIDIA will have nothing to do with GPU chiplets, and AMD knows that.


In a recent interview with VentureBeat, Sam Naffziger, AMD senior vice president, corporate fellow and product technology architect, talked about GPUs consuming up to 600W of power by 2025, and much more. But then VentureBeat’s last question was a doozy: “Do you feel that compared to NVIDIA and Intel, we’re in a state of divergence when it comes to design, or some sort of convergence?”

Naffziger replied, explaining: “It’s hard to speculate. NVIDIA certainly hasn’t jumped on the chiplet train yet. We have a big lead there and see great opportunities in that. They will be forced to do so. We’ll see when they deploy it. Intel certainly jumped on that. Ponte Vecchio is the poster child for chiplet extremes. I would say there is more convergence than divergence

Naffziger continued: But the companies that innovate in the right space are the fastest to get a head start. It’s when you provide as much of the new technology as what the technology is. Whoever is the first to innovate has the advantage

Now, if you’re aware of all this… I reported in July 2017 (5 years ago now!) that NVIDIA was rumored to be switching to multiple GPUs on future GeForce graphics cards… more about that here† It hasn’t happened yet, but AMD is rumored to have multiple GPU chiplets on its upcoming Navi 31-based graphics cards, but NVIDIA will be stuck with a monolithic GPU design for Ada Lovelace.

VentureBeat’s previous question — As for the concern that we’re hitting walls with things like Moore’s Law hitting boundaries and other physical limitations looming, how concerned are you right now? — had Naffziger reply with: “I am concerned in the sense that it stimulates new dimensions of innovation to achieve efficiency. Silicon technology isn’t going to do it for us

AMD on NVIDIA does not use GPU chiplets: 'we have a big lead there' 01 |

He continued: “We’ve been seeing this coming for a long time. As I said, the delivery times are long. We’ve invested in things like the Infinity Cache, chiplet architecture, and all of these approaches that take advantage of new dimensions to maintain profits. So yes, it’s a big concern, but for those who prepare in advance and invest in the right technology, we still have a lot of opportunities