🤣I find it HIGHLY unlikely that AMD is going to drop EPYC Zen 6 full core count down to 96 cores.
I also find it IMPOSSIBLE to believe that AMD will reduce the core count on their EPYC dense. The entire purpose of the dense version is for cloud computing that doesn't utilize the L3 very well and where max core count rules.
I am thinking this is a little too far off the side of reason as well.
It's just my speculation at this time, but I find it much more likely that AMD will lean into chiplets and IOD technology in the next generations of processors. This will lead to smaller die sizes and more complex combinations of IOD's.
This gives AMD the best combination of cost reduction and scalability.
I believe a 32c CCD is technically possible, but it may well be economically inferior. Think of it this way....
AMD can EITHER create fewer VERY expensive CCD's and have a less complex packaging having fewer IOD's OR they can create MORE less expensive CCDs and use more IOD's.
I see no down side to the scaleable approach and many advantages.
Scaling via an IOD that has 4 channel memory allows AMD to scale memory channels along with processor cores keeping the system in balance.
Each generation AMD gets to decide what memory technology they will support the platform on, select the number of cores in a CCD that can be fed from that memory, then create an IOD that matches the number of channels needed to the number of CCD's supported.
Scaling of cores then becomes something that is only limited by the socket power and max memory channels supported by the socket design.
These kinds of decisions are my daily bread and butter. From a business standpoint, this looks pretty appealing to me.
I even doubt that there is a performance hit by putting fewer cores on a CCD and having more IOD's vs having fewer IOD's and having more cores on the CCD as work is scheduled by thread in the OS.