Yeah, I wasn't sure what this "fundamentally broken" thing that couldn't be "ironed out in firmware" was about either. I think it was conflating the 32 TLB thing that got tossed around a few months back on Twitter with the more recent "unpatchable" security vulnerability clickbait stories. I don't think TLB issue was ever something that would be resolved by firmware, I believe the idea was that properly optimized apps wouldn't hit it. The speculation was that Apple might do a larger TLB because that might be easier than waiting for apps to be optimized. All of this stuff needs to be taken with a big grain of salt.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Silicon Mac Pro Configurations: Everything We Know
- Thread starter MacRumors
- Start date
- Sort by reaction score
Yeah, I wasn't sure what this "fundamentally broken" thing that couldn't be "ironed out in firmware" was about either. I think it was conflating the 32 TLB thing that got tossed around a few months back on Twitter with the more recent "unpatchable" security vulnerability clickbait stories. I don't think TLB issue was ever something that would be resolved by firmware, I believe the idea was that properly optimized apps wouldn't hit it. The speculation was that Apple might do a larger TLB because that might be easier than waiting for apps to be optimized. All of this stuff needs to be taken with a big grain of salt.
That is not how Apple support policies work. Apple is not going to support Mac model X because different model Y released a new version. Once Apple stops selling model X year A then that model goes on the Vintage/Obsolete countdown clock. Other Macs getting updates does almost nothing to stop that clock from counting down.
If Apple did a new Mac Pro with Intel then the MP 2019 model would be on that countdown clock too along with the overwhelming vast majority of other Intel Macs. Once 75-90% of the Intel Macs have gone over the obsolete cliff macOS on Intel is likely going to be in an almost comatose state.
The Mac Pro is likely a less than 100K/yr product (decent chance substantive lower than that as Apple Silicon transition has progressed. ) . It isn't going to be able to hold back millions of other Macs from going over the "obsolete" cliff all by itself. The notion that Apple will provide longer support for folks who paid lots more money ... there is zero in Apple policies that says that at all. If Apple did a refresh it would be more so for the customers who tend to squat relatively long periods of time on systems (and macOS versions).
If Apple had done a Mac Pro Intel update in 1H '21 they probably would have rode it for two and then quit. Similar to how the Intel Mini drifting in a narrow niche now and Apple is going to close that door within 6-9 months. (or sooner).
Apple support policies give some leeway on when Vintage/Obsolete statuses get invoked. 5-7 years. With the transition for the Intel systems there is no slop in the number. Hit 5 and done. So the fact that Apple has stopped sales on pragmatically the whole Intel line up means millions of those Intel systems are going over 'the cliff' each year going forward.
Apple cut the MP 2013 quicker than 5 years after retired but that was in large part they went into Rip van Winkle mode on an update. Any super small volume Mac Pro Intel model would be in a similar boat of being so far off track with overall Mac update norms that the exception would likely generate an unusual 'chop' rather than something that runs so far off the product line up strategic plan. At best, there would be some super narrow macOS code maintenance just for that last gasp Mac Pro ( no new drivers , no new features , very little new software. etc).
[ once all the the pre T2 Macs go over the cliff, the hack-my-mac-past-Apple-support-window is likely going to get much harder. So backport hacks is probably going to be less smooth. ]
Yeah. So my question was, how are you supposed to connect more than two Max chips together, as the OP was proposing. If the answer is "use Max chips that aren't Max chips" then that seems like both an answer and a non-answer.
Wow this is the first time I've seen this written out clearly. Everyone always tries to argue it's just software issues. Sounds like a dumpster fire for Apple. $5k Mac Studio Ultras are flawed pretty badly... When does Apple start dumping these flounders? Is class action possible for a performance issue like this?
Sounds more like a "software development" issue. Not sure developers have figured out how to leverage the type of the GPU (aka multiple-cores) on the AS chips yet.
The Mac Pro is on display in flagship Apple Stores (e.g. downtown/central location in a metropolitan area), but they are still not available to order in store. You would think they would keep at least the base model on hand. The price is still ridiculous, especially considering Apple is only offering about $1500 for a trade-in. You could buy one and give it back to them at a $4500 loss...
Yet, when you actually look into it, it IS that software developers have not adopted practices that use the tile-based approach that Apple’s GPU’s use.
Now, this is a bit of a quandary. Should Apple make it easier to adopt the current approaches and therefore guarantee that developers will not play to the hardware’s strengths?
To me, Apple is using the M series era to make a clean break based on their future roadmaps, but there is a fine line between requiring total rewrites (that WILL be worth it in the long run) and supporting current approaches. Both have pros and cons.
I hope that's the case too. Apple has said before that the Mac Pro is modular by design. They created the current expandable model based on feedback from target customers, such as the motion picture industry. I can't imagine those same customers would want a closed box Mac Pro that is like a higher-end Mac Studio. They want PCI-e slots.
It better be modular. Upgradeable storage and RAM. Better not try to make it mini at the expense of functionality
Epistemology is not really required learning for internet "journalists".
So far, it looks as if the only thing preventing storage upgrades on the Mac Studio is that the CPU needs to be re-configured to accept different sized flash modules. Even if you can't upgrade at least the modules can be replaced like-for-like without ditching the whole logic board. Unlike RAM, flash only allows a finite number of writes, so soldering it in is pretty unacceptable (I mean, in theory it should still outlast a mechanical drive - but then you wouldn't solder a mechanical HD in either...)
The actual setup - 'raw' Flash modules with the SSD controller on the SoC - is very similar to the 2019 Mac Pro (raw Flash with the controller on the T2 chip) for which there are storage upgrade kits and a suitable configurator tool.
If the new Pro uses the same approach as all the other Apple Silicon chips to date then the RAM can't be upgradeable. LPDDR5 RAM doesn't come in plug-in form - it is designed to be surface mounted as close to the CPU as possible (on the CPU package in the case of Apple Silicon) to keep the connections short, fast & power efficient. Certainly the M2 Max-based configurations suggested in the "rumour" article would presumably be LPDDR5-based.
OTOH, if Apple wants to offer 1TB+ RAM configurations to replace the Intel Mac Pro they'll have to add regular DDR5 slots somehow - but that would need a more conventional SoC design.
The twitter thread about limited TLB size doesn't make a lot of sense.
The claim is that a 32MB TLB is enough to sustain the M1 Max's GPU performance, but not the M1 Ultra's... except that the M1 Ultra is two M1 Maxes (Maxs? Max's?) so it has twice the TLBs.
So how can one TLB of a particular size be enough for one set of GPU cores, but two TLBs aren't enough for two sets of GPU cores? Hmm.
Not really. They can simply add RAM slots to the system.
Have fast on-package RAM the way they do with all current M1 chips, and users can put slower RAM in the slots.
Pretty easy to do from an engineering standpoint.
Thing is, all those high-end GPUs that are so much better than the M1 Ultra will happily plug into a bog-standard Xeon or Ryzen tower where they will happily run Windows or Linux software optimised for x86 and DirectX, OpenGL, OpenCL, CUDA etc. Do Apple really want to compete with Intel/AMD on making commodity workstations if their only selling point is MacOS and a shrinking handful of applications that have no Windows or Linux equivalent?
Apple could make a "more conventional" Xeon-like ARM CPU with loads of PCIe, external CPU support and external DDR5 RAM and it would probably beat Xeon in terms of how many cores you can fit on a chip and performance vs. power. Trouble is, power consumption - ARM's party piece - is a big deal for laptops and SFF systems and is also becoming a big deal for high-density computing and servers. However, the Mac Pro is none of those - it's not portable, has plenty of space for cooling (and multi CPUs if you want more cores) nor is it designed to be farmed in vast numbers that cost a fortune to power and air condition. Low power is still "nice to have" but not such a killer feature - especially at the cost of being able to run x86 binaries. In any case, those big discrete non-Apple-Silicon GPUs are still going to guzzle power and belch heat.
Then there's the development costs issue - the rumoured Ultra/Extreme SoCs have the great advantage that they're basically made by linking together the same Mx Max dies that Apple can sell in far larger quantities in MacBook Pros and Studios (and the even larger-selling Mx Pro is essentially "just" a M1 Max with a chunk of GPUs left off). An ASi "Xeon Killer" with PCIe replacing half the Thunderbolt controllers and 1TB-capable regular DDR5 controller would probably mean developing a whole new die.
The point about the M1 Max/Ultra/Extreme may not be what they can't do - i.e. beat a high-end $$$$ GPU at OpenGL/CL - but what they can do - i.e. eat ProRes or anything else supported by the media engine for breakfast while sitting in a small-form-factor housing.
So I think I'd still put my money on a Mac Pro either being little more than a rackmount Studio Ultra or some sort of scalable system taking multiple MPX-like Ultra/Extreme "compute" cards. It seems like better use of the existing technology. They could always keep the Intel Mac Pro alive for a few more years.
I sort of like the idea of Apple hook’ing up with AMD for one last Intel compatible Mac Pro. Imagine a custom Thread Ripper with 128 cores. AMD would be so excited they’d probably give Apple a huge discount on their chips for the business. Then maybe revisit getting rid of x86 in a decade in 2032 or so when the M chips are very mature.
Yeah, I agree with you that Apple likely doesn't want to invite comparisons with other workstations, so they will keep selling their on-package "unified" memory as if it's magic. (It is PR gold, really. How many people think that M1/M2 Macs are faster than PCs only because of on-package/unified memory?)
As for what we're likely to see in an upcoming Mac Pro, I suspect a dual-socket system.
I doubt the M2 Max will deviate much from the M1 Max, i.e., it will have a single connector along an edge to allow it to combine with another M2 Max and form an M2 Ultra, but that's it.
It should be easy enough for Apple to put two M2 Ultras in one system. They will need some kind of interconnect fabric, but it would surprise me if they haven't been working on such a fabric for years already.
MacOS will also need some modification to ensure that processes stay local to one chip or the other as much as possible, but that's all pretty straightforward.
Yeah but Apple is trying to move away from x86 as fast as they can. Must be a pain to support both ISAs with all their software.
As long as Apple is still selling Intel/T2 Macs they're pretty unlikely to launch a new MacOS without Intel/T2 support. But once the last Intel Mac is removed from sale then it's feasible that the very next MacOS will drop all Intel support.
Until then, your particular Mac may still get dropped but that's not inevitable, esp. if its a T2 (I'm sure non-T1/T2 machines will be the next to go) - but when the Intel code is gone from the OS, it's gone...
AFAIK the "5 years after last sold" and "vintage/obsolete" is only about parts and service & is required by consumer law in some jurisdictions, which doesn't entitle anybody to new software features not promised when they bought their machine.
I would have expected M2 Max 12 CPU 40 GPU cores, M2 Ultra 24 CPU 80 GPU cores, and M2 Extreme 48 CPU 160 GPU cores. With the next M2 Studio getting the first 2, and the M2 Mac Pro the last 2.
Hard, but probably good for the software ecosystem in the long run to easily support different architectures. Plus all the work is really done isn't it. They have had the Intel port for over a decade. It seems like cruising along with it for a few years doesn't seem that challenging. Hell, NetBSD supports like 100 different CPU architectures!
Last edited:
The new Mac Pro wheels will roll around your workplace automated like Roomba. Going from desk to desk as needed.
If you can't find it, it's in the breakroom. Or the restroom....
But seriously, it's nice to see them cover all the bases with super high end product offerings. Successful companies run the gamut. And with the Studio fitting nicely in between, no one is left out. They'll sell plenty to Hollywood for sure.
If you can't find it, it's in the breakroom. Or the restroom....
But seriously, it's nice to see them cover all the bases with super high end product offerings. Successful companies run the gamut. And with the Studio fitting nicely in between, no one is left out. They'll sell plenty to Hollywood for sure.
What if one of the ARM-64 CPU makers (such as under license from Qualcomm) were to begin offering a DIY ARM ATX desktop motherboard that could at least run "bare metal" Windows ARM-64, as well as potentially dual-boot macOS? Thus allowing for a Hackintosh running on ARM?
Or maybe: an ARM-64 CPU equipped PCIe card that would allow running macOS while booted under Windows x64. Sort of how Parallels can now run Windows while booted under macOS.
Last edited:
There has been mention of Apple using unpublished calls for their ARM instruction set. And, if they have the license we think they do, that’s fine as long as they support the entirety of the ARM instruction set while they do. That means that it’s quite possible that macOS wouldn’t run well OR at all on non-Apple hardware.
Didn’t they have something awhile back where you had to scavenge from a Mac (I think it was a ROM chip?), BUT you could then run Mac software on it… maybe it was just a dream?
I have A LOT of apprehension with the rumours of a reduced size [volume] MacPro. The whole point of the professional end of the market is to use specialist PCI-e cards like sound cards, capture cards etc.
You can’t fit those in a smaller unit. So it would effectively be a Mac Studio in a tower format.
The MacPro 2019 is a really an industrial design masterclass. The collaboration and engineering that went into it were unprecedented for Apple. To turn around and provide one that can’t be used with all the professional hardware would seem incredibly foolish.
You can’t fit those in a smaller unit. So it would effectively be a Mac Studio in a tower format.
The MacPro 2019 is a really an industrial design masterclass. The collaboration and engineering that went into it were unprecedented for Apple. To turn around and provide one that can’t be used with all the professional hardware would seem incredibly foolish.