GPU options like that would mean too many completed mobo SKUs to stock...
Apple will most likely segment the APUs by overall core counts, the more "powerful" the Mac is supposed to be, the higher the core counts on the SoC...
Updating my speculated APU line-up...!
M1 - 4 P / 4 E / 8 GPU / 16 Neural Engine / 16GB LPDDR5 / two USB4 ports
MacBook Air / 13" MacBook Pro / Mac mini
M1X - 8 P / 4 P / 16 GPU / 24 Neural Engine / 64GB LPDDR5 / four USB4 ports
14" MacBook Pro / 24" iMac / Mac mini
M1Z - 16 P / 4 E / 32 GPU / 32 Neural Engine / 128GB HBMnext / four USB4 ports
16" MacBook Pro / 27" iMac / 30" iMac Pro / Mac Cube
There is no 'Mac Cube" coming.
Also, it is likely getting nothing new in Neural Engine across the line up either.
If Apple cranks up the P and iGPU cores then they'll also need to crank up the system cache to keep all those cores fed. The Neural engine is plenty fast for inference on data and the scope is pretty narrow with respect to application usage.
The large shared "System Cache" and Unified Memory is where Apple is getting a chunk of their performance gap from ( moving the data closer to the computional engine that is using it and fewer copies of data) . It also works better because the overall SoC die is small. As push into much larger die they'll loose that if they don't cover more data to deep the 'cores' feed. Pragmatically that means not going to to get "more" of everything.
The System Cache is also centrally located in the whole set up. The the center of the die only can be so large. Start to get core block groupings too far from the center and some of the "secret sauce" goes away. If going to another block of 4 P cores and/or 4-8 Core GPU block then probably not going to be room around the center for another Neural block even around an incrementally bigger System Cache center. (also need to let the Memory controller(s) more access to the System Cache. So basically has 2-3 more "access" consumers of a fixed resource. The Neural , Image Processing, Secure Enclave ,and the rest of the uniformly standard Apple Silicon stuff probably keep a constant size "seat" at the incrementally bigger table. )
As for USB4 ports ... not need if the iGPU isn't crippled to just 2 DisplayPort streams. Probably means Apple would need a better embedded Thunderbolt controller. It only stays USB4 is Apple propagates the kneecapped controller out to the rest of the line up. ( they may.. that would be cheaper for them. )
System Memory on HBMnext.... likely not.
M1T - 32 P / 6 E / 64 GPU / 64 Neural Engine / 256GB HBMnext / eight USB4 ports
Mac Pro (Threadripper-sized package)
No Threadripper sized package coming either. 8 USB4 ports only if added a MPX card. The base system probably will not go past the 4 ports the MBP maximizes out at.
Apple isn't likely to chase the ThreadRipper, EYPC , and/or Xeon SP packages on core counts and maximum number of memory controllers ( and DIMM count).
Same here with the HBMnext. Regualarly in the triple digit GB probably should have ECC support. ( and something past the internal only DDR5 stuff.) If it just for the tower this should be back in the DIMMs zone.
Apple discrete GPU (Lufika) - 64 GPU / 64 Neural Engine / ?? HBMnext ??
There is not really good indications that Lufika actually is a discrete GPU. Lufika could be pragmatically be an integrated GPU. Perhaps chiplet. Perhaps using a small amount of HBM for "back side", local cache. ( something akin to AMD's new "Inifinify Cache" ) That might drive the package size out to Threadripper size but for different .
The 32 (or up) GPU above for the above coluld very well be the Lufika implementations. To go to that many GPU cores Apple would need to do something to pull more of the System cache 'workload' into a larger GPU only cache for shared local space between the very large number GPU computing function units. Effectively a subsystem cache.
Plus seriously need a better location for Display output ( the 2.5 stream limit of the M1 is low . 2 DisplayPorts and a touchbar screen ) is pretty limited when Intel Xe-LP is at 4 streams and MAD is larger also.
Mac Pro may have DDR5 DIMMs implemented as a secondary RAM array, up to 1TB...?
The application cores on "seconardy" RAM isn't likely to happen at all. Nor is Apple likely to scale up to 1TB . Something like the 600GB range ( or lower ).
All desktops will have two USB-A ports & a HDMI port...?
Errrrr, probably not . The iMac and Mac Pro aren't going to pick up system embedded HDMI. The standard configurations of the Mac Pro will get one defacto with the discrete card. Put attached to the side of the box. Probably isn't going to happen. With 4 TB ports if someone wants HDMI they can just get an adapter if running solely off the iGPU in the Mac Pro. Most Mac Pro folks though are going to have a add in card. The add in card will come with the HDMI. Apple might allow a option to buy a Mac Pro with an iGPU only. ( because user is going to buy their own or use it in a more server setting where don't necesarily need one. ). Howver, the standard configs will likely be like the current ones and come with a MPX card .
Type-A ports will probably stick around the more pressing issues with the desktop SoC will be whether Apple allows/provisions 10GbE or not more uniformily. ( and still go to two on Mac Pro and perhaps bring the iMac Pro up to two also. )
