AMD RDNA 3 Radeon Graphics Feature Chiplet Architecture

This is not a drill: Chiplets are coming to Radeon GPUs
AMD’s upcoming Navi 3x (RDNA 3) graphics are moving from 7nm down to a 5nm process, but these will be far more than an enhanced die shrink. Rather, this next chapter in the Radeon story could mark the biggest shift in modern 3D graphics acceleration – at least from an architectural standpoint. The reason? Chiplets. Yes, CHIPLETS in a gaming GPU.
We have, of course, grown accustomed to AMD’s chiplet approach to CPU architecture since the company introduced Ryzen processors five years ago, but multi-chip GPUs will be new territory. And it’s crazy to think that this radically different approach hasn’t just been announced for some time in the future, but is “expected to launch later this year” according to AMD.
As shown on the above slide, AMD will be bringing a re-architected Compute Unit and pipeline enhancements to RDNA 3 graphics, plus a “next-gen” Infinity Cache, in addition to the headlining move to a chiplet approach rather than a large monolithic die. AMD claims that these improvements, along with the move to a 5nm process, have the potential to boost efficiency by greater than 50% over RDNA 2.
As to the overall roadmap, AMD showed a slide indicating that after the upcoming RDNA 3 (“Navi 3x”) products at 5nm, the next generation RDNA 4 (“Navi 4x”) will be on an “advanced node”, though it was not revealed whether this will be an enhanced 5nm or a smaller process tech.
For more information on these and other announcements you can view AMD’s press release (link), and follow this link to AMD’s Financial Analyst Day for access to all of the presentations.