There is ALWAYS a market for the a higher end, all powerful Mac. It is a small market, but Apple has shown that they understand now that they have to provide "something" to accommodate those users.
Except, in the past, Apple has been able to build a pro system very easily using existing chipsets developed by Intel and AMD (or IBM & Motorola if you go way back). Although the Mac Pros of the past have had some nice Apple refinements that didn't develop themselves, most of the heavy R&D had already been done by the CPU and GPU makers, who were serving a far larger market from which to claw back their development and tooling costs. Designing and making small batches of bespoke chips is
insanely expensive.
With Apple Silicon, Apple have been able to start with cores designed for the massive iDevice market, develop them into the base Mx chips for the huge low-end MacBook and iPad market. Then they use more of the same cores to design the Mx Max die - and
that serves everything from the 14" MacBook Pro up to the Studio Ultra (The Mx Pro being basically a Mx Max with a bit left off, the Mx Ultra being two Mx Max dies) Of course it's not quite as simple as "cut & paste" in practice but it is a lot more cost effective than producing three unique die designs - and spreads the cost over the entire Mac and iPhone line.
...but its looking as if the Mx Ultra is as far as that is going to "scale" so to produce something equivalent to the 2019 Mac Pro in terms of RAM and PCIe capacity they're going to have to produce a complete new die design - probably a separate CPU and GPU if they want to out-CPU-core AMD - and also ditch the unified RAM concept that is a significant factor in M1/M2 performance. That's a whole new level of R&D investment c.f. previous Mac Pros and the
only tangible way of recouping that investment is via sales of Apple's lowest-volume product. Now, I'm not saying Apple
won't do that, but its a much bigger risk than previous Mac Pros that would be very hard to justify via some sort of "halo effect" for the rest of Apple's range.
Also, they
could go with AMD for graphics (and/or re-build some bridges with NVIDIA) but then, on many tasks, the expensive new Mac Pro will only ever be as good as AMD's latest, and just the same as a commodity x86 tower. Or, they could do something more radical of their own - which would then likely give so-so performance on existing code optimised for PC GPUs. That against the background of a lot of rather conservative Mac Pro customers for whom the only reason they're still paying a premium for Macs is that they can't afford to disrupt their workflow to adopt new software.
What PCI cards exist that would work in an Apple Silicon mac? Given how few Macs have slots, it is not surprising that no one bothers making cards Mac compatible.
List here (PDF):
https://www.sonnettech.com/support/downloads/manuals/TB_PCIe_Card_Compatibility.pdf - M1/M2 compatibility does thin down that list somewhat, but there are a number of supported cards. if they work in a TB-to-PCIe expander there's no reason why they wouldn't work on an Apple Silicon Mac Pro.
There's really two issues here - general PCIe cards (for storage, I/O, specialist AV interfaces etc.) and high-end GPUs. I guess it wouldn't be too hard for Apple to re-purpose some of the 6 TB4 ports available on a Mx Ultra to provide 4 PCIe lanes each, which would be enough for many I/O and storage purposes - but (even if Apple U-turned and added driver support for AMD GPUs on Apple Silicon) the sort of high-end GPUs that people are pining for need 16 lanes apiece for full performance. The Xeon W in the 2019 MP has 64 lanes of PCIe. M2 Ultra really isn't the tool for that job.
Also, what happened to the parallel processing approach from a few decades back? A university networked a bunch of G4s together and got the fourth fastest supercomputer in the world? That inspired Apple to start making blade servers for a while
The idea of clustering is not dead and still widely used - see: folding@home for an extreme example...
Thing is, though, Apple doesn't really have a dog in the high-density computing race any more - and I think it was primarily the rise of Linux & Apple's switch from PPC to Intel that did for that (...one Intel blade server running a Unix-like OS is as good as any other - there's no advantage to a nice friendly UI or a shiny aluminium case).
However, CPUs in those days were mostly single core - even the 2006 Mac Pro used two dual-core Xeons to get a quad core setup. Nowadays, we have 20-core M1 Ultras and AMD are making
64 core Epycs - and its generally more efficient & easier to program a bunch of cores sharing the same RAM and I/O.
Still, with Apple Silicon, looking at clusters of (low power) Mx Ultras might be more productive than trying to pretend that it's a Xeon/Threadripper replacement suitable for a big box'o'slots.
As for the Mac Studio question...
The only real issue is that the new M2 Pro Mini, when fully tricked out, out-performs the M1 Max Studio -
especially the base Studio with the 24-core GPU - and costs about the same. That problem goes away as soon as Apple upgrade the Studio to M2 Max and add a bit of inflation to the price (not nice, but probably inevitable). The Studio is still "nicer" than the Mini in various other ways - support for an extra display, front-facing ports and what looks to me like better internal construction (replaceable, if not upgradeable SSDs, all external ports on replaceable daughterboards etc.). Apple would be nuts not to update it eventually - but it's par for the course that they're dragging their feet (and the MBP is obviously a bigger selling product and will be first in the queue for M2 Max chips). The Studio always did leave a clear gap in the lineup for a Mx Pro desktop at around the price of the higher-end Intel Mini.
The problem arises if Apple tries to kludge together a M2 Ultra "Mac Pro" that is really just the equivalent of a Studio + PCIe enclosure, and tries to force people into it by not upgrading the Studio. I wouldn't put it past them, but I hope not. A "real" Apple Silicon Mac Pro would be in a different price bracket and not really compete with the Studio. Or, they could admit defeat on the Cheesegrater/Big Box'o'slots concept (AMD/Intel is the tool for that job) and just rename the Mac Studio as Mac Pro - then market the hell out of it to all those new prosumer content creators out there in the blogosphere who should be the ideal target market for a plug-and-play FCPx appliance.
appleinsider.com
