The popular opinion currently denies their entry into the next-gen RDNA 4 lineup, but it still doesn’t mean that they are entirely canceled. What I mean by this is that it is possible that these GPUs are postponed because of a few problems.
According to some recent reports, it is currently not feasible for AMD to produce such high-end cards based on their new die design. If you remember from one of the recent videos from Moore’s Law is Dead, Tom talked about a leaked diagram of how the next-gen GPUs will be designed based on the Navi 4C architecture.
Initially, AMD was planning to launch 4 GPUs based on Navi 41 and 42 dies and these dies supposedly feature a multi-chiplet design with somewhere between 13 to 20 chiplets.
Unlike, the Navi 31 die, it is almost 3-4 times more chiplets on a single die. Now, even though we have a few theoretical diagrams from the leaker Everest of how they would look like, another Twitter user called creper9000 made a full render of these chips.
It can be clearly seen that the die is divided into 3 zones where each zone features 3 Shader Engine dies on each Active Interposer Die, having their own infinity cache and GDDR memory controllers. Moreover, it also has a dedicated I/O die which is connected to every AID and apparently, the design will also incorporate die stacking.
So, definitely a big improvement in the die design but as of now it is unclear to why exactly it is not possible for AMD to include such a design in the next-gen lineup. Because at least the die stacking is already being used in Datacenter dedicated MI300 GPUs.
Whatever the case is, we have to wait for AMD’s official announcement regarding these rumors but we can clearly see that AMD still has a chance to make a comeback if it wants to.
