I don't know what other dies they would have planned, but it would be a little odd for the X700 part to be the top die. Obviously there's no rule that says AMD has to stick to the conventions they've used before, but they usually do.
There's also a matter of what they do with dies that aren't fully enabled. There are still going to be some defective parts (and large GPU dies make this a lot more likely) and those that need some hardware disabled to hit voltage targets. This seems to ignore that these exist at all.
With that much potential for variation it's hard to say how they'll handled the naming, but I wouldn't be too surprised if we see a departure from the current scheme to something that's better suited for a multi-chip approach.
I'm also not even sure that games would benefit all that much from such a massive number of shaders. We've already seen how much performance scaling starts to fall off after the 3080 even at 4K. The raw performance is probably great for compute workloads, but I don't think we'll see 2.7x when it comes to gaming.
- Fair on the naming scheme, we may see a radical departure from the old convention. All the same AMD has had a hard time sticking with an actual naming convention for much longer than a few gens, it'd be nice if they planned their current naming scheme in anticipation of things like chiplets etc. Won't hold my breath though.
I figure defective dies will end up as the non-XT mid tier parts, so a 7900 non-XT for example would be a 2.5 N33 (one die with half CUs disabled) so we end up with something like 200cu/12,800SP.
If this gen has shown anything, its that there is a hunger and money for high performance parts and AMD wants some of that pie. Also, 8K gaming appears to be the next resolution battleground (which covers VR as well) and this level of scaling up would uniquely benefit AMD's performance in Ray Tracing given their brute force approach to the technique.
Chiplets would likely allow AMD to be much more nimble in meeting demand as well. No need to do a bunch of wafer allocation, if there is demand for their ultra high end, just send more chiplets through that packaging channel. If not, pump more dies to the mid-range. No more wafer outlay and shortages because they made too much of Chip A and not enough of Chip B.
I wonder what happens below N33 though. Some of that market will be fed by APUs (rumors of all Ryzen processors getting an iGPU abound) and discrete add in boards can just be rebrands of N23/24 or something.