I do not necessarily agree with this. In order to announce an M2 Ultra Mac Pro, I think they need to address two questions:
- Is the M2 Ultra fast enough for a Mac Pro class of machine?
- What justifies the existence of Mac Pro when the Mac Studio is also available with the same SOC?
It very likely wouldn't be the same SoC. The Studio has both the Max and Ultra. Pretty doubtful that Apple is going to put a laptop Max package in a Mac Pro. ( especially at the $6K price point that they have moved the Mac Pro entry price to. ) . The bulk of Mac Studio sales are going to come from the Max variant not the Ultra.
So really only really talking an overlap of higher end BTO Ultra SoC. And doesn't necessarily have to be exactly a M1 physically identical Ultra. Same SoC but which one? Something optimized for laptops or something constructed for desktops (e.g., add in PCI-e backhaul provisioning).
The iMac Pro had over x16 PCI-e v2 lanes that went nowhere. The Mac Studio either could have a SoC that had x32 PCI-e v4 lanes that went no where. Or make a slightly smaller package and just no add on lane provisioning in a smaller package.
There is a persumption here that if core count is X than package size is exactly the same as the M1 era. That isn't necessarily true. if Apple can have two dies in a package it isn't a big leap to have 3 and keep the core counts exactly the same.
Let's think about the answers from Apple's point of view:
- Apple has announced first-generation Mac Studio as a significantly more powerful machine than the Mac Pro, which is actually true, considering the key areas that Apple is focusing on: video (Final Cut Pro), music (Logic Pro) and coding (Xcode). If they can position the M1 Ultra as a more powerful solution than the Mac Pro, then obviously this is even more true for the M2 Ultra, let alone an overclocked version. By the logic of Apple-marketing, M2 Ultra is not a disadvantage of Mac Pro, but and advantage of Mac studio.
I think this a bit overblown on "significantly more powerful". Apple was very explicitly careful to carve out a subset of the MP 2019 space to just cover the most commonly bought CPU and GPU components ( 16 CPU cores and W5700X ). Also didn't pick a far above average RAM data working set footprint problems to benchmark against either.
The Mac Studio covers more performance ground than a iMac 27 model from 3-5 years previously. But that is similarly true about the rest of the line up versus their previous Intel models.
There is a small bit of "you don't have to wait for the Mac Pro if this system fits" there. But yes it is a general long term trend across whole PC market where can move many workloads down to 'smaller' systems over time.
- Mac Pro was always about modularity and expandability (well, except for the Trash Can). Having PCIe slots for storage, networking and audio acceleration, is a difference big enough for Apple to justify the existence of Mac Pro next to the Mac Studio. Even more so if Apple adds third-party (AMD) GPU support, not to mention if they announces the rumored compute modules.
Provisioning PCI-e slots has nothing to do with CPU or GPU core counts. The CPU and GPU core clusters that Apple uses as die construction building blocks don't 'do' PCI-e
at all.
The Mac Pro having PCI-e slots can simply be the case that the Mac Pro has the PCI-e backhaul PCI-e controllers hooked up and the Mac Studio does not. Has nothing to do with core counts, and everything to do with different functionality value-add. Multiple internal drives ( that has value to end users ). More I/O configuration control with 3rd party products ( that has value to end users ).
The Apple Watch and HomePod Mini has the same SoC , but the stuff attached to the SoC is different. AppleTV and iPhone.. same SoC ... different stuff attached to the outside.
In fact if look at Apple's entire line up of Silicon packages ... which ones go into one and only one product? At this point none. Not a single one.
That appears that it will slightly change for a while if the XR Headset has two SoCs in it. Pretty good chance there a highly customer package in there. But reportedly Apple is busy at work already for another headset that probably will change other components than the SoC to construct a difference. And the price point on the XR Headset is way out there also.
At the most extreme Mac Pro BTO cofinguration there will be probably be another corner case , but mild, exception that the SoC is so expensive it isn't in another product. But also a pretty good chance the building block chiplets inside are used in the 'lower half' SoC "Ultra with real chiplets" that is shared elsewhere outside the Mac Pro product line.
A Mac Pro with M2 Ultra, (without user accessible RAM and without third-party GPU support)
3rd party GPU support has diddy-squat to do with internal PCI-e slots. The M1 and M2 series SoC support over 50 PCI-e cards through Thunderbolt. Apple could have added eGPU support any time over last 3 years if wanted to go that route.
If want PCI-e slots the FIRST thing need is some decent PCI-e backhaul provisioning. Period. Obsessing about drivers for a very narrow subset of cards
completely misses the point. If Apple adds 3rd party GPU drivers and all stuck with is x4 PCI-e v3 bandwidth internally for all of the internal slots , that isn't going to make many folks happy.
is a slightly lower-end machine then the 6,1 MP was in 2019, that's not even a question.
A MP 2023 with x16 or x32 PCI-e v4 internal backhaul and 6 internal slots is way better situation tha nthe MP 6,1 was in 2019. 6 slots would be
more than the 5,1 had (aggregate bandwidth than 6,1... bigger than that too. If two x16 PCI-e v4 controllers the aggregate would be the same as the MP 2019 model). How is that even remotely on 'equal footing' with the 6,1 in 2019; let alone worse footing ?
But I believe it's existence is still fully justified using Apple's logic.
The problem with M2 Max (and any Ultra that excessively just pairs exact twins) is that M2 Max is bloated in size compared to the M1 Max. The "Max" was about as big as could make a die and still fit inside the parameters for using TSMC's InFO-LSI packaging technology. Is the bloated M2 still inside the limits... maybe not. There is another more expensive option of CoWoS-LSI , but it is more expensive. If Apple was using the M2 Ultra where the Ultra package was optimized for the Mac Studio and most highest reuse of only the laptop die to lower costs... then somewhat doubtful they would be going to more expensive packaging. One of the top priority objectives would be controlling costs.
Pretty good chance though that Apple's logic is to change the package ( even if it gets slightly more expensive. Apple's price mark up on M1 Ultra's isn't exactly frugal. Should be money there to cover incrementally higher base costs ) so it can be used in multiple products instead of just one.
www.newegg.com
The Puget has arrived. I will be unboxing it later, but it was 85 pounds and feels like a bigger machine than the 2019 Mac Pro by far. I'll do a side by side size comparison once I get her out of the box later. I have a friend helping me move the piano up to the music studio in a bit since I want to keep the puget in my production area in the house across from the Mac Pro. 
