Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Adding GPUs could happen, but upgradable RAM is impossible while keeping the crazy memory bandwidth of Apple Silicon. Removable RAM is dead. Can't have those speeds with slots.
 
In my opinion, Apple may not introduce the "final" Mac Pro until the M3 generation SoC's with their 3 nm process is unveiled late this year. Two reasons: 1) less heat-related issues and 2) the M3 is supposed to get a new memory controller better-suited for multiple CPU and GPU cores.
That sounds reasonable - a more expensive and bigger MacStudio with no discernible benefit would really be strange
 
Adding GPUs could happen, but upgradable RAM is impossible while keeping the crazy memory bandwidth of Apple Silicon. Removable RAM is dead. Can't have those speeds with slots.
Yes - but you could have a slower type of memory and an intelligent controller handling that.
 
The real problem with the Apple chips is that they do not do virtualization. On my Studio, I had hoped to be able to run Parallels with Windows subsystem for android, but that is impossible.

They don't do virtualization? Then how is your Parallel App running on top of Apple's Hypervisor framework? M-series does support virtualization.


"...
The Hypervisor framework has the following requirements:

Supported hardware
The Hypervisor framework requires hardware support to virtualize hardware resources. On Apple silicon, that includes the Virtualization Extensions. On Intel-based Mac computers, the framework supports machines with an Intel VT-x feature set that includes Extended Page Tables (EPT) and Unrestricted Mode.
..."

All the virtual machine software on Apple silicon run on top of the Apple's hypervisor framework. The different virtual machine vendors can add 'features' on top but the foundation is commonly shared. (for better or worse. Contributes to why VMWare is lagging because because they have to loose their own lower level but try to look like they didn't (so seamless with rest of VMWare VM empire. The other empire is their primary business. )


"...
The Virtualization framework provides high-level APIs for creating and managing virtual machines (VM) on Apple silicon and Intel-based Mac computers. Use this framework to boot and run macOS or Linux-based operating systems in custom environments that you define.
..."


Apple does some basic stuff for macOS and Linux layered on top of the hypervisor. Minimally completing with the VM vendors. Apple isn't chasing Windows here though.


Android on Windows on MacOS needs nested virtualization; not just plain virtualization. This hasn't been a high top priority for Arm more than several years back.

"...
Before the release of Armv8.3-A, it was possible to run a Guest Hypervisor in a VM by running the Guest Hypervisor in EL0. However, this required a significant amount of software emulation, and was both complicated to implement and resulted in poor performance. With the features added in Armv8.3-A, it is possible to run the Guest Hypervisor in EL1. With the features added in Armv8.4-A, this process is even more efficient, although it still involves extra intelligence in the Host Hypervisor.
..."



Arm v9 this is more commonly 'baked in'. It didn't make much sense for Apple to do nested virtualization before Arm had worked out the kinks on what would be the standard way to do this. The issue is more so where does Apple 'merge in' with the Arm standard approach with the specifics of Apple's security model and their implementation. It would help Apple long term if they did something that didn't hugely deviate from the standard Arm approach.

It wouldn't be surprising if Qualcomm ( what Windows on Arm is generally built toward) isn't on the bleeding forward edge here either. Ampere Computing (and other Arm Neoverse implementers probably are).

It is much better risk management for Apple to get a solid "one level" virtualization done before moving on to multilevel. However, if going to pragmatically ban everyone else's hypervisor at some they do need to add another level to a rock solid foundation.
 
  • Like
Reactions: DavidSchaub
That’s because they went mobile first and scaled up. Mobile doesn’t need virtualisation. My M1 Max is great but I can’t help thinking the architecture is a dead-end for desktop. The Studio is essentially a headless MacBook Pro.

There really wasn't a very efficient Arm model for nested virtualization until quite late in the Armv8 evolution.
It isn't just "mobile first". There is a standards adoption rate factor that plays a role here.

The Studio's Ultra isn't really a 'headless" MacBook Pro since and Ultra wouldn't fit in a MBP.

What is missing more so from the Studio is higher bandwidth I/O that would lower performance per watt a bit. The M1 Max ( and presuably M2 Max ) have like x4 PCI-e v4 lanes for general purpose I/O. That is actually less I/O than a AMD/Intel laptop chip. Actually less than Intel's new N-series "Celeron" replacements.



An Apple chip with a UltraFusion connector(s) could be augmented on I/O. The problem with the M1/M2 Max is that it is a bit too monotholic focused. But it could be disaggregated without 'blowing up' the basic approach to add in a bit more I/O.

The nested virtualization is something that could be weaved in on M3 (maybe M4) generation iteration when have the basics done. Given the M2 Max can host 96GB it could be used there too. ( Virtual Machines aren't going to be completely foreign to all of the upper 10% percentile MBPs with large memories. )
 
Last edited:
So, following on from @Boil , @DavidSchaub and @Alan Wynn, here’s a model for an ASi Mac Pro:

  • big case, massive cooling, even more than the Studio and it’s copper. This means more power consumption, and thereby higher clocking of ASI CPUs than other classes of Mac.
  • Lots of PCIe slots because they, together with a fast processor, justifies a Mac Pro for a small but significant market - because pci has more bandwidth than thunderbolt. And for those who need GPU expandability…
  • Apple GPU expansion cards, similar to MPX. This is a bit of a stretch but conceivable. The bigger challenge here for graphic artists is support for the right software / APIs…
  • ASI memory: ssd and memory integration is so fast on ASI that a half-full 4tb internal ssd gives you 2tb of virtual memory, which is largely indistinguishable from main memory. This won’t satisfy everyone but will work for almost all workstation tasks. The really big work is done is done in a cloud anyway. You could possibly add more memory, if Apple enabled adding RAM disks via an expansion card or pci slot, but that seems unlikely to me.
This might work and I am hopeful. Much is arguable. We will see.

However I suspect we greybeard computer enthusiasts may need to move beyond the model of “my big computer needs internally expandable memory and storage because that’s what I was doing in 2005, or 1995”. I’m not sure that’s the optimal model for desktop computing any more. Buy a system that will work for you for 2 or 3 years then change it seems to me to be a more effective approach not just for 90% of users, but 90% of power users.
 
Apple is getting incredible performance on the M series by having massive bandwidth between CPU/GPU and RAM.

That just can't be done with PCI GPU cards and/or RAM sitting in some kind of upgradeable slot, so just forget those two options.
Our company owns many high performance DELL PC workstations - they very rarely if ever get RAM upgraded or PCI cards added. And RAM upgrades, if they happen, are sometimes problematic causing machine instability.

It's like losing the DVD drive - people realize they never used it anyway.

It all depends. If you are into audio mixing it's much neater to have some madi pci card in your machine. If you are into video you want your kona pci as well. And the most obvious is GPU upgrade right.

When it comes to optical drives I still have use of BluRay to burn BDXL4 discs from time to time. It saves space when it comes to storage.
 
Adding GPUs could happen, but upgradable RAM is impossible while keeping the crazy memory bandwidth of Apple Silicon. Removable RAM is dead. Can't have those speeds with slots.
If upgradable RAM is impossible, then so is the GPU because it requires the same crazy memory bandwidth. Apple’s GPU is not like other GPU’s where the GPU RAM is segregated from main RAM.
 
In the 8 year I used my G5 cheese grater I did mid-life upgrades.
In the 10n year I used my trash can I did mid-life upgrades.
Expecting to use the next gen MacPro for a period of 8-10 years again, I do hope I can do mid-life upgrades again. If not, the ROI on the initial purchase doesn't work out.
 
It's a straw man that doesn't even exist yet. Let's all panic!

"Based on nothing other than some past mistakes, and because I have no real information, let's consider the worst thing that could happen."
"Did you hear that? Apple might do the worst possible thing."
"They're almost certainly going to do the worst possible thing. They did it before!"
"I can't believe it. Apple did the worst possible thing AGAIN?"
"I know, right? They're doomed.

Rinse and repeat.
Also don't forget: "Steve would've never done this!!"
 
  • Like
Reactions: b17777
If upgradable RAM is impossible, then so is the GPU because it requires the same crazy memory bandwidth. Apple’s GPU is not like other GPU’s where the GPU RAM is segregated from main RAM.

iGPU in SoC handles display output, utilizing UMA RAM...

ASi GPGPU(s) handles compute/render tasks, using on-card RAM...
 
  • Like
Reactions: DavidSchaub
If you work in an industry where things can change rapidly or change on a daily basis, the need to be able to upgrade you computer to adapt to those changes is extremely important. This is why you will find many many businesses using Windows desktop PC's because they can be upgraded. There is always the scenario where a company wins a new contract but finds their current desktop computers cannot handle the requirements of the new client. Many clients will have very strict deadlines that need to be met and thus it would be disastrous for a company to lose out on new business because their desktop computers can not be upgrade and they are not able to afford brand new updated spec'd desktop computers.

If Apple will not allow their desktop computers to be user upgradeable then simply put many many businesses will not buy them.
 
I suppose Apple could build a Mac Pro which allows you to swap in and out CPU+RAM+GPU packages. Bit of a 'hefty' upgrade - maybe they will offer some kind of trade in service? Or a subscription :-/

The M1 is built to be entirely self contained so no idea how else they'll do it.
 
i know this site is literally called mac rumors but can people just chill out

no use arguing about speculation

the mac pro is currently apple's weakest product, i'm sure they know it

who knows what they've got cooking. if they did get M2 extreme to work, would you be interested? The amount of time they're taking hopefully means they're not going to f*ck this up.

Who knows, maybe the wheels will have brakes this time.
 
How are they surprised? Apple & Nvidia burned all bridges so we'll never get the state of the art GPU hardware+software. AMD is doing ok, but is severely lacking in the software library department, even on Windows.

The only option Apple has is to make their own powerful GPU. It will be horribly expensive to make something that's very competitive or even way better, and Apple would still need to write some very decent libraries for existing ecosystems (pytorch, tensorflow, etc) or hope that someone will do it for free. (hasn't happened yet) There are still very, very few apps that make use of al the special silicon in the new chips, like the 'neural engine'.
 
In the 8 year I used my G5 cheese grater I did mid-life upgrades.
In the 10n year I used my trash can I did mid-life upgrades.
Expecting to use the next gen MacPro for a period of 8-10 years again, I do hope I can do mid-life upgrades again. If not, the ROI on the initial purchase doesn't work out.
What do you use your Mac for? I bought a tricked out Mac Studio that I write off on two clients. Are you sure you need a Mac Pro? I mean I would like to upgrade as well, but it has always been a non-issue for pro users.
 
Since the Ultra chips were not able to scale the performance of the individual chips, and since these are really niche machines that sell in such small volumes, I think Apple decided that they could not justify investing the resources required to recoup their investment on such dedicated chips for these niche machines.

I am wondering if Apple will instead do what Intel did with their Xeon chips (or still does, I do not follow Intel anymore) and just use complete dual chips on their machines instead of custom fusing them together as they were said to do with the Super Duper Ultra Supreme Deluxe XXL chips for the new Mac Pros.
 
  • Like
Reactions: slash8slash1
Honestly, everybody would love a stylish case for 19" units. Make it so it can be used both vertically and horizontally, built-in fans and basic power supply and I/O and you just pop in your Apple 19" unit with a buch of M2 units, add 3rd party units for storage, networking or whatever special needs you have and round it off with (overpriced) Apple designed covers and wheels.

Then again, I don't know a good use for a high performance Mac, Logic runs fantastic (and completely noise free) on an M1 Macbook Air, Final Cut is too buggy and lacks features to be used professionally.
 
Why don't they have both Intel and Apple Silicon options?
Apple wants control over the important parts of their products and it allows them to depend on specialized silicon like machine learning cores. Apple's GPU design is significantly different then Nvidia and AMD, so they very likely don't want to keep supporting AMD chips longer then necessary. If Apple continues AMD support on the new Mac Pros, it will probably be constrained to non-display work (non-realtime rendering and compute).
 
Last edited:
  • Like
Reactions: DavidSchaub
Really sounds like Apple is making a high-end computing appliance without a market. They’ve done something like this before with the 2013 Mac Pro and it flopped. It will flop again this time.

There is no point in competing in this end of the market without modularity and a high degree of component replaceability and expansion.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.