Good point. When developers write software for macOS, do they take into account the unified memory model of Apple Silicon? Or is this stuff just taken care of by the compiler? If the latter, then perhaps it is just laziness on Apple's part. My impression was that it would require extra work from developers to support two different models of GPU operation, but perhaps not.
The deprecated Intel code paths obviously still work, but all AS code to date has been able to assume unified memory. Could an AS MP with a PCIe GPU run this code unmodified? I guess in the worst case it would just require some kind of translation layer / shim. Most apps aren't going to use much VRAM anyway.
What's the marketing implication? If the next Mac Pro could accept PCIe GPUs, surely that would just remove criticism from certain high end Mac users? The majority of Mac customers would just continue buying MacBook Airs, oblivious to the change. It might give Studio Ultra customers pause. But then the Mac Pro is £3K more expensive and a completely different form factor, so maybe not.
Given its mobile roots, the whole Apple Silicon thing is performance / watt. All their products are thin / compact, with silent cooling. A big box with a power guzzling GPU is at odds with this aesthetically, so sits awkwardly in the range. The current Mac Pro form factor is 100% Xeon workstation. Apple commonly reuses case designs over an architectural transition, but the v2 might change things up. Or maybe the v1 will just hang around for a while as a bridge for those with PCIe cards, then get discontinued. Given the ease with which they could have upgraded the logic board to an M3 Ultra, if 2025 ends with no MP revision, it'll likely suggest the latter.
Again, though, is there much market for a Mac Pro that doesn't take NVIDIA cards?
I agree about the hubris. Apple is also very image conscious; their whole persona is based on 'perfection', so they almost never admit error.
No argument there.