The thing is, the Ultra is made up of two Max dies connected by a "silicon interposer" called Apple UltraFusion. The UltraFusion is capable of linking four Max dies, which is why the rumors of a "twice as big as Ultra" Extreme chip exist in the first place.
Errr who openly stated that a single UltraFusion connector was capable of linking four dies?? There are lots hand wave Rube Goldberg proposal that slap alot of "miracle occurs here" magic that claim it could possibly work. However, most just generally sweep substantive problems with their 'solution' under the rug. I've seen most of these 'rumors'.
When the Ultra M1 came out there were more than a few grumbles that there was non-linear performance increase. That is just with one connector between two dies. Add 3 more dies and 3 more bottlenecks you think it is going to get better scaling? It probably won't.
Back in the orginal rumors there were
Jade C ( eventually M1 )
Jade C-chop ( eventually M1 Pro )
Jade 2C ( Max and Ultra )
Jade 4C ( rumored extreme)
Jade and Jade C-chop were not the same die at all. Same with Jade 2C (it has very high design overlap with Jade C-chop). . So why is Jade 4C necessary the indicative of being the exact same die as Jade 2C.
All can pull away from that is that Apple stuffed some die dies together. The necessity of Max really is not there at all. And the performance evidence of just doing two is indicative that very specific interface won't scale much further. (Make no mistake is a very significant achievement to 'glue' those two together and get almost invisible NUMA impacts and more than decent performance. But zero evidence that there was gobs of latent
unused bandwidth there. )
The reality is that the basic Max type die is not a particularly good 'chiplet' . It is very chunky and once weave in the placement issues with the Memory models very awkward to scale past that one interposer connector without taking some lumps on performance (and NUMA transparency).
Your plan of an "incrementally bigger" chip would be a larger undertaking, because you'd be designing an entirely new chip,
Aha... if the 4 chip solution
always was yet another die design therein lies the problem.
Frankly doing an incrementally bigger chiplet isn't a big problem. Apple already did that with the M1 Pro and M1 Max. The Max was designed to be a bit bigger. And yes the Max was a bit more expensive than than the Pro . It is not the case of whether it is going to be an more expensive chip to make it is a matter of controlling how much more expensive. Because spinning out of control on costs and/or ruining Perf/$ is a major show stopper.
If have split the Max into a chiplets can reuse one of those in multiple pairings.
CPU-I/O chiplet + mini GPU chiplet
CPU-I/O chiplet + full GPU chiplet
CPU-I/O chiplet + too big for laptop GPU chiplet ( Ultra )
CPU-I/O chiplet + full GPU + full GPU ( more than Ultra but probably too big for Mac Studio )
[ and possibly CPU-I/O chipset + full GPU + CPU-I/O chipset ]
three products all using the same CPU chiplet ( so shared volume cost savings into the MBP market)
The major problem would have been to saddle the chip to just the Mac Pro with dramatically low unit volumes.
If do an even bigger functional decomposition of the I/O from the CPU. If don't have the 'too chunk' chiplet then can get more shared reuse across a wider set of SoC products.
Once break out of the mindset that needs to only have a chunky, monolithic "Max' chip optimized for laptops there are more options. But that is entirely not what the M1-M4 where. So "Extreme" coupled to that mindset made less and less the more Apple showed the M1-M4 progressions.
even if it was an evolution of an existing design. A doubled Ultra chip would effectively be "free" from an engineering standpoint. The problem is that nobody knows what the market is like, and any failure rates at the fab (it's never zero) would be hugely expensive.
Nobody knows what the market is? Isn't there a whole department inside of Apple called "Marketing". Nobody in that group has any clue nor spends any time researching what the market demands are? I suspect the folks who work on 'spin' and 'branding' are in that group to and likely are a bit clueless, but nobody in the whole department? Probably not.
Perfectly accurate insight into the market future? No. However, no knowledge at all as the what the market is? Somebody isn't doing their job at multiple levels of the organization.
More likely Apple got some 'educational' lessons out of doing the Extreme. Probably leads to a better mindsets toward approaching doing a chiplet deliberate design. Also probably helped flush out what are currently doing with AI server chip.
The real problem was the notion of the Extreme was Apple doing a SoC that would be just used in only one product. The R1 in the Vision Pro is a similar boat anchor drag on cost management. ( I'm sure there was some hand waving Apple Car R1 excuse that faded away with time. )
To be blunt I don't thin Apple has a really good handle on the problem of churning the Max chip of the MBP 14"/16" into the trash can every 12 months either. So not particularly surprising not M6 Pro/Max in addition to no Ultra that generation.
