4900 for the for base 24 core/56 gpu with 6 slots please.
Apple currently charges $3,999 for a Mac Studio with 20-CPU , 48-GPU core M1 Ultra. If the M2 goes to multiples of 12 on CPU count then same thing 24 core CPU , 56 GPU (out of 72) is likely where the M2 Ultra Studio would go. The 'full' Ultra Studio is $4,999 ( $1000 to add 16GPU cores ).
Apple has already set the price levels for an Ultra powered Mac. Price cuts below that for M2 generation and forward are pretty unlikely. TSMC fab manufacturing costs for more advance nodes going up reinforces that. Making these multiple chip packages isn't going to be cheap (even if Apple shifts to incrementally better chiplet designs for the desktop offerings). Apple has not significantly used the M-series transition to dramatically lower Mac system prices.
Bigger power supply, PLEX switch , more aluminum for bigger case , drilling fancy holes in the case , slots , 'low volume tax', etc .etc. Likely not getting any lower than the current $5,999 for the Mac Pro. Perhaps one way they'd get back to the Studio price price would be to serious kneecap that entry Mac Pro in some way. (e.g, 512G SSD or something silly like that. There are folks who 'hate' Apple SSD drives so if trying to sell some boxes to them who are going to buy and put their own drive in the "too low" capacity isn't as much of a hurdle. They are ignoring it anyway. Lots of folks grumbled at the 512GB SSD configuration. If Apple didn't sell more to the 'haters' with the MP 2019 than they thought they would, then it is probably gone. If enough folks bought it perhaps it says around. ).
With the M2 Ultra there is decent change that Apple would do away with the kneecapped SSD capacity at the $5,999 price point. The Studio and Mac Pro would baseline entry capacities of 1TB.
And option of apple gpu cards.
If Apple canceled the M2 Extreme because of too low volumes to be interesting and opportunity costs are too high ... then the likelihood of discrete Apple GPUs is pretty slim. That is an even more expensive niche to branch out into.
I think there is presumption that Apple would make something priced like a Nvidia 4070, 3090, or AMD 6800 . Probably not. The W5700X which the internal Ultra SoC beats is priced at $1000
https://www.apple.com/us/search/Radeon+Pro+W5700X+MPX+Module?src=aos_alp
If Apple was going to make something that was going to take the place of a W6800 ( $2,800 ) and W6900 ( > $2,800 ) , then they probably would price them around the same magnitude. All the Mac Pro 5,1 and 7,1 folks who ran out and bought "way cheaper than official Apple GPU product " cards aren't really going to motivate Apple much to make a GPU card. Those folks didn't buy the other Apple offerings, so why would they buy these new ones. Hence, land in just about the same low volume boat as the M2 Extreme with the same opportunity cost overhead (if not higher).
If the Ultra package is giving the user a 'no extra charge' W5700 then that $6,999 MP 2019 is now a $5,999. If the M2 Extreme had come through a similar play where the W6800 costs would be folded into the baseline.
IMHO the larger disconnect is the sizable number of folks who are price anchored on the maintream mid-high end consumer GPU card prices and mapping that into Mac Pro configuration prices. Apple has been disconnected from that for very long while. A subset of those cards "happened to work" inside of a Mac Pro 2019 , but that never was Apple's primary point with doing the Mac Pro 2019. Those "happen to work" cards all got an almost free piggy back rides out of the primary work that Apple wanted to do. Intel's CPU and chipsets needed some more modern UEFI updates so those got weaved into the MP 2019 , but that was doing most of the motivating and primary work there.
Apple dumped UEFI with M-series. Changed the kernel security model. Haven't signed any 3rd party GPU drives in over 2 years. Haven't even incorporated an GPU driver abstraction the drive object hierarchy. Even if Apple did a dGPU it would be useless in the general PC market having completely different boot firmware and radically different drivers. That means it won't get any the R&D cost amortization over a large volume , which means the units costs will be largely detached from mainstream GPU prices.
The notion that "well Apple does iGPU and those will make it cheap". Cough. go look at how much money Intel is burning getting into the dGPU market. AMD 7900 benchmarks don't match the spec hype from a couple of months ago and now 'targeting' the 4080 instead of 4090. (not quite the same scale of driver meltdown but substantive issues ). Likewise, Nvidia isn't king kong of the iGPU market either. Extremely high quality, broad spectrum drivers in a different space aren't cheap.
There is just far more higher synergy across the iOS/iPad/WatchOS/macOS/realityOS ecosystem for deeper Apple iGPU app optimization then anything some dGPU tangent is going to generate.
[ Scoped down from the complexity of a display GPU API (Metal ) to a much simpler compute only API ( OpenCL , SYCL , or perhaps "compute only Metal" ) which also throws out boot issues is a more tractable path to weave in a compute accelerator for more 3rd party compute grunt work. It wouldn't need to be an Apple GPU foundation. Not necessarily super consumer commodity market cheap (not going to make the lowest price anchored folks happy), but far more aligned in objectives and pricing. ]
