MP All Models Intel may be the only way to save the Mac Pro

[deconstruct60]

Interesting article by macworld: https://www.macworld.com/article/1518344/mac-pro-apple-silicon-transition-intel-xeon-processors.html
Personally I do not see this happening. – If for no other reason, it would simply be Apple losing its face.

I think this is where the article goes off the rails.

"...
That’s plenty fast, of course, but compared to the current M1 Ultra Mac Studio, it’s not worth anywhere near the $17,000 starting price for the top-of-the-line Intel Mac Pro with a 28‑core Intel Xeon W processor. When the Mac Pro launched in 2019, its closest competitor was the 18-Core Intel Xeon W in the iMac Pro, which paled in benchmark comparisons (13,453 versus 26,604 in Geekbench multi-core tests). Granted, the Mac Pro cost about $10,000 more than the iMac Pro, but that sizable price gap only underscored how much more speed you were getting. ..."

The 'Oh, it is only $10K more' notion is a problem for the Mac Pro.

The price points on those 24 and 28 Intel W-2200 processors were whacked back in 2019. Intel's ">1TB RAM capacity" tax was a dubiously money grab back that probably drove more folks to AMD faster than it made Intel more money over the long term. Apple piling on top to slap their tax-on-tax to fill their pockets too , but also seems to have partially lead them into a long term ditch also . ( Oops the Extreme might be too expensive to sell the 2023 completitive market. )

Here is a comparison chart I used in a pervious post for W2200 , W2300 , W3400/2400 . It is all 24 cores. 28 is even more ridiculous.

W-3265M / W-3345 / W7-2495X / W7-3455

Recommended Customer Price

$6987.00

$2751.00

$2189.00 - $2199.00

$2489.00

Intel product specifications

Intel® product specifications, features and compatibility quick reference guide and code name decoder. Compare products including processors, desktop boards, server products and networking products.

ark.intel.com

If only really needed 256GB or less of RAM it really shouldn't have been $10K more. Even just taking the 'M' off the end drops the price to $3694 ( ~$3K less). The big change from 2018 ( when Apple probably conjured up the current Mac Pro prices ) and 2023 is that AMD is the dominate player in setting the price points now; not Intel.
And when drop down into the 16-18 core zone there is also a whole different level of price competition.


Apple needs to be far more concerned about putting value into the Mac Pro price points than in engaging in the notion of 'an extra $!0K sprinkled on top is worth it'. A 24-28 core equivalent Mac Pro that is inccrementally faster but $3-4K cheaper will have much higher Pref/$ . That is very much more an issue now than it was in 2018-2019.

HP z4 and cohorts in the workstation market are likely a much better threat to the Mac Pro now than they were in the 2019.

Intel isn't necessarily going to bring the biggest Pref/$ win here. Apple isn't really set up to play in the maximum core count and most expensive BTO configuration war . Intel playing that game is in denial also. ( e.g., hiding the accelerators of the 2400/3400 dies behind even more expensive markups. ) Apple has to shine a lot of light on their accelerators and not over price them.

 

[Reggaenald]

Wow... just wow.... The amount of disconnect here is just.... wow

First, Apple publicly went on an apology tour and admitted that the 2013 trashcan was a failure, so they released the 7,1 to make up for that.

Furthermore, look what Timmy did in 2012:

https://techcrunch.com/2012/09/28/t...pple-maps-points-to-competitive-alternatives/

Stop drinking the kool-aid

So? What else could they do? What good did. their apologies of a bad designed product do to the people that bought that exact product long before they admitted to anything but “can’t innovate anymore my ass”. When you design a consumer product, you should think about how this product changes the consumers life for the better.
An apology some years after the fact doesn’t remedy anything that happened before. They should have know how bad their design was. But they didn’t.
Different example.
What am I to do if they “apologise” for the 2016-2019 MPBs? Do I get a machine that won’t maybe randomly fail? A better keyboard? More value for the money? Anything?
No.
If an apology comes after the fact and is only good for customers of the next generation of things, it’s not an apology. It’s a realisation of mistakes of which they don’t have to be the bearer.
Their customers are.
There are things that never should have been released.
Apple only admits to what the market thinks, maybe.
You can keep the kool aid. I have enough problems with my underperforming 2k MacBook “Pro”.

 

[alfogator]

The reason I like MP and cMP is that i can add cards, isn't that the limiting factor for arm,does not let you do that or is just apple arm that does not allow in cards?

school me!

The arm mac can use cards, there's no inherent limitation: I have an egpu enclosure that I use with pcie cards slotted in (network, storage). As long as there's a driver it will work.

Apple doesn't make an arm mac with builtin slots but they could if they wanted to.

 

[olimerkido2]

I wonder what the truth is regarding the "one more Intel Mac we're excited about". Was there going to be a refreshed Intel Mac Pro at some point? Now would be another decent opportunity given the new Xeon chips, but it feels like the time has been and gone.

In an ideal world though, it would be great to see an update. Tweaked motherboard, new Xeons, larger base SSD size, and a couple of new AMD MPX modules. Just quietly refresh that at the same time as releasing the Apple Silicon version.

It would help keep any negativity about expandability of the AS version at bay e.g. "users who still need 1.5TB of RAM can still purchase the Intel version". That'd buy Apple some time to work out how to really get huge performance out of Apple Silicon for the next version.

I think actually said more Intel Macs, meaning plural, could have just been referring to multiple configurations of an Intel Mac, anyways did happen, was 27 inch iMac got refreshed one last time during the Apple Silicon transition with a 10 core CPU I think was.

 

[olimerkido2]

The reason I like MP and cMP is that i can add cards, isn't that the limiting factor for arm,does not let you do that or is just apple arm that does not allow in cards?

school me!

Is not a limiting factor of ARM, not adding cards is common in ARM because ARM is usually used in phones and laptops, the amount of ARM desktops other than Apple's current ones is near zero, but ARM can support expansion cards if Apple chooses to allow you.

 

[Unregistered 4U]

Yeah but what else could they say really…. I have no doubt they have workshopped chip designs with zero (or few) gpu cores so they can use dedicated gpus in the Mac Pro - either instead of or along with the integrated ones. For all we know they have their own AS-based gpus on the table with all (or mostly) gpu cores. considering the kind of work (and money) that goes into these things and what they are able to charge for them at the end we might all be way off the mark with our low expectations of what they can achieve on a AS Mac Pro. As far as RAM goes they could just make it moot by shoving so much in there that no one would ever complain.

Oh, to be sure. Absolutely anything could happen. Based on all of the Apple Silicon systems that have been released, though, they’ve all followed the same pattern. If the future Mac Pro also follows that pattern, no one should be surprised.

 

[AlixSPQR]

Why would Apple continue with x86 code in macOS for such a small number of people? I can't see that happen.

 

[Confused-User]

The more important indicator as it relates to GPU’s, though, is more about what direction Apple is providing to developers regarding “How to develop for Apple Silicon systems”. And, that direction has been “no GPU’s other than the integrated GPU”. A lot of what Metal on Apple Silicon is and does (and what Apple likely wants it to be) depends heavily on the GPU sharing memory with the CPU, which would not be likely with anything other than an SoC solution. Until Apple announces a change as to how they handle GPU’s (say, at a future WWDC), the only GPU any Apple Silicon system will have is the one on the SoC.

That's an interesting thought, but you're missing an intermediate possibility, which is that Apple could hide all the complexity of a multi-*Apple*-GPU setup in the OS. I recently read about an Apple patent which, surprise, has some applicability to this (it deals with divvying up workloads among multiple GPUs).

I think you're entirely wrong about unified memory being unlikely for multiple GPUs. I think it's the heart of their software paradigm and they will maintain it regardless of how many GPUs are in the system. One interesting question is, will they build enough smart NUMA awareness into the OS so they can hang gobs of RAM off any additional GPUs? I think they will, but we won't know until they announce something.

 

[Confused-User]

Surely you jest😱

Lou

Surely I don't. I don't think they'll build any more x86 Mac Pros, but if they did, they would definitely be wise to use the latest EPYCs rather than anything from Intel. How is that even debatable at this point?

 

[Confused-User]

My question is simple:

Why would anyone even purchase a Mac Pro, even if it has Apple Silicone?
[...]

I dunno. Even Bros need laptops. The novelty value of a computer with fake boobs might sell some.


Can we please learn to spell that word correctly?

 

[BellSystem]

What a ding dong this guy is. Apple isn’t going to keep Intel around for a single model. The reality is the biggest customer of the Pro is audio production and ProTools is in beta with native support. As soon as it drops all the studios will eat these up.

 

[apparatchik]

The M3 Extreme has to be the key, along Thunderbolt 5 (120Gbps/PCIe 5) support.

It wont reach 1.5TB of RAM but 512GB memory on the unified architecture plus 8-10 lanes of TB 5 should allow for a compelling and expandable machine.

 

[oneMadRssn]

UMA is about a single memory space for CPU and GPU that allows pointer-passing instead of copying data. It doesn't require soldered RAM chips.

But standard DDR5 DIMMS have the memory controller on the DIMM itself. Whereas on AS the memory controller is on the SoC. So no matter how you slice it, making AS compatible with user-upgradable DIMMs would require either proprietary Apple-only memory modules or Apple relinquishing the memory controller to be off-SoC. I don't think Apple nor Apple users would be happy with either outcome.

 

[mattspace]

Why would Apple continue with x86 code in macOS for such a small number of people? I can't see that happen.

they maintained the Intel codebase from NeXTSTEP, through to the first Intel release of MacOS X for literally zero people (except perhaps Steve, who loved his Thinkpads).

 

[oneMadRssn]

Apple's chips are clearly designed for mobile products first. The Mac Pro in general goes against the "AS philosophy". I hope we don't have to wait too much longer for Apple to release something. At least we will be able to discuss an actual product on its merits instead of speculating wildly to no gain.

Obviously it's indisputable that the M chips are evolved from the A-series mobile chips, which prioritized energy efficiency above all else. However, I think this feature pays dividends in desktop uses too - less active cooling, smaller power supplies (which are an annoying point of failure), better packaging.

While I doubt that Apple is going to retire the Mac Pro name, I think the new Mac Pro will be more like the Mac Studio and less like the pre-2012 Mac Pro. If I was placing a bet, I would bet against it having user-upgradable RAM. I bet it will it have soldered NANDs for the primary storage, but maybe it might have M.2 slots or something for user-addable secondary storage (in addition to TB4 and such) or an active cache or something like that. Finally, I bet it will not have a user-upgradable GPU, but I can see it having quasi support for external compute modules such as eGPUs (not for driving displays, but accessible for workload).

 

[Confused-User]

But standard DDR5 DIMMS have the memory controller on the DIMM itself. Whereas on AS the memory controller is on the SoC. So no matter how you slice it, making AS compatible with user-upgradable DIMMs would require either proprietary Apple-only memory modules or Apple relinquishing the memory controller to be off-SoC. I don't think Apple nor Apple users would be happy with either outcome.

This is completely false. I can't even imagine what you might have read that would lead you to believe this. Memory controllers first mirated from the "northbridge" to the CPU when AMD introduced the Athlon K8 (2003), and when Intel (eventually) introduced Nehalem in 2008, it became pretty much universal (leaving out some corner-case RISC chips like the Alpha, which had them back in the 90s, and a few really oddball cases like Power8 that left them off the CPU). Perhaps you're thinking of registered or buffered DIMMs? They still don't have memory controllers.

Apple's large chips will definitely continue to have their own memory controllers, and this says nothing at all about whether they could support DIMMs (or even, argh, possible Apple-proprietary memory expansion cards).

 

[Confused-User]

Obviously it's indisputable that the M chips are evolved from the A-series mobile chips, which prioritized energy efficiency above all else. However, I think this feature pays dividends in desktop uses too - less active cooling, smaller power supplies (which are an annoying point of failure), better packaging.

That's true, but it seriously misses the point when we're talking about large processor designs!

High-efficiency cores are a *HUGE* win for large multiprocessor chips. Have you ever wondered why the large chips from Intel and AMD (60- and 64-core models, for example) have such low "base" clock speeds? And why their boost speed is still under 4GHz (typically closer to 3.5GHz)? It's because they simply can't push enough power through the chip, and then cool it enough, to run all the cores at high clocks.

Apple's CPU cores are in many ways *ideal* for large-core-count chips, because of their great efficiency. If they could manage to get 64 of those cores into a single package (say, two or four chiplets, with an IO hub to link to GPU chiplets, or just for lots of PCIe), even the M2 cores would likely stomp current x86 chips. Of course, that's *if* they do this.

There are three things that might stop them from going big like this. First, the financial - it may be unattractive to invest enough to build such large core count chips. As I've said elsewhere, this is a reasonable argument, possibly countered by a need for such chips by future products like xR devices. Second, technical - I can't readily put my hands on the size of the individual cores, or of clusters including L2$, but getting eight of those on a chiplet might be too much. (I think not, though, for N3B, just from looking at the die shots of the M2.) Also, they'd have to grow up their uncore, especially their NoC, a *lot*. This doesn't seem like an insurmountable challenge though.

The third barrier is simply intention. They may not want to. That, I can't address.

 

[Spaceboi Scaphandre]

I can bet that Apple can't wait to get rid of x86 binaries inside the Mac OS as soon as it can be done with least amout of pain possible.

Writer of that article is high on hopium. We will never see another Intel Mac.

I doubt the x86 binaries are gonna get removed entirely. Rosetta 2 will still have it's purpose, especially since it's available in ARM Linux distros now too.

Rosetta 1 got removed because PowerPC was a dead architecture. x86 ain't dead anytime soon

 

[Unregistered 4U]

That's an interesting thought, but you're missing an intermediate possibility, which is that Apple could hide all the complexity of a multi-*Apple*-GPU setup in the OS. I recently read about an Apple patent which, surprise, has some applicability to this (it deals with divvying up workloads among multiple GPUs).

There’s already multiple Apple GPU cores, all accessing the same set of RAM as the CPU. And this fits with what Apple’s been communicating to developers.

I think you're entirely wrong about unified memory being unlikely for multiple GPUs. I think it's the heart of their software paradigm and they will maintain it regardless of how many GPUs are in the system. One interesting question is, will they build enough smart NUMA awareness into the OS so they can hang gobs of RAM off any additional GPUs? I think they will, but we won't know until they announce something.

They already use multiple GPU’s cores. They don’t currently (and are unlikely to) use any GPU’s that don’t have access to the CPU’s main memory. There’s nothing I can say that will make ANYONE conclude against what they strongly believe, but physics IS a thing. And, the GPU bandwidth of even the slowest M1 is at the high end of what PCI is capable of currently. Could is a powerful word and pretty much anything “could” happen. However, based on what Apple’s shown about their SoC strategy, some things are definitely more likely than others.

 

[Spaceboi Scaphandre]

Interesting article by macworld: https://www.macworld.com/article/1518344/mac-pro-apple-silicon-transition-intel-xeon-processors.html
Personally I do not see this happening. – If for no other reason, it would simply be Apple losing its face.

On the other hand they showed to be quite pragmatic about reversing some of their decisions in the past and even admitting they were wrong. The Mac Pro family with the 2013 model and follow up 2019 model are excellent examples.

IMHO they could have updated the Intel Mac Pro with a newer Xeon chip some year(s) ago, there were even rumors about an XCode beta featuring newer Xeon ids in its code. But I assume from a marketing pov it would simply not have been wise – benchmark comparison of the M chips vs. the Intel Mac Pro would not look so nice anymore…

Personally I would still love to see another Intel Mac Pro revision – sooner or later I am sure it will join the Apple Silicon family – but for the time being I would most likely buy another Intel model to be able to run Windows and macOS on the same machine natively, upgrade PCI cards, RAM and disk space.

Does the Mac Pro even need "saving?" The M1 Ultra Mac Studio kicks all kinds of ass, even the fully specced out 2019 Mac Pro's ass at a massive fraction of the cost.

I know you guys want swappable upgradable parts and got PTSD of the Trash Can, but Apple Silicon is no slouch and showed you can have an uber powered professional workstation that a fraction of the size. Sure expansion will be little to none, but Apple Silicon lasts a stupidly long time so in the long run that might not matter.

 

[Confused-User]

They already use multiple GPU’s cores. They don’t currently (and are unlikely to) use any GPU’s that don’t have access to the CPU’s main memory.

I completely agree - I've been saying that all along.

There’s nothing I can say that will make ANYONE conclude against what they strongly believe, but physics IS a thing. And, the GPU bandwidth of even the slowest M1 is at the high end of what PCI is capable of currently. Could is a powerful word and pretty much anything “could” happen.

PCI would be a *really* crappy interconnect if they were doing things the "traditional" way. In fact I'd guess it would be infeasible, even PCIe5. If they do put GPUs on PCIe, I expect them to have a customer interconnect, much like NVLink or Crossfire.

However, if they do hang big gobs of RAM off each GPU, and go all-in on NUMA, then... what? Perhaps something closer to hypertransport? As you say, physics is a thing. Getting all the chips close enough to each other for this to work well would be a very interesting challenge.

However, based on what Apple’s shown about their SoC strategy, some things are definitely more likely than others.

I don't think that they've shown their entire hand yet, so I don't think that that's a meaningful statement. After all, before the M1 Pro and Max shipped, there was nothing in the M1 that suggested that they would. Before the Ultra shipped, nothing suggested that it would. Well, except for the ultrafusion interconnect we could see in the die shots, but that didn't tell us much.

 

[mateo14]

but Apple Silicon is no slouch and showed you can have an uber powered professional workstation that a fraction of the size.

I get confused when I read many similar comments.

Professional?

Ok, I use a wired keyboard and mouse because I'm lazy.
I have two hard drives, a SSD, and I plan to add another SSD.

It's hard to imagine connecting four hard drives, a DVD drive, a keyboard, and a mouse to Mac studio. It won't be an esthetically pleasing and organized desktop.

I like to put everything inside the case of a computer because electronic devices shouldn't be treated a saucer for my cup.

Size?

I have a 55-inch monitor on my desk, so I don't worry about the size of my computer. It will always be much smaller than my monitor, and I can accept it.

Of course, I also have an old server with HP-UX (50 kg) on a window sill, which is a much heavier computer than Mac Pro.

Why do some users think that modern computers should be a size of a book?

Some computers should have bigger cases if we don't want to have bunch of cables on our desks. Please, correct me if you know a better solution.

I have two older Macs: Mac mini and iMac. It's difficult when you have to replace something in them. They are fragile, but they can be fixed in many situations. I can't say it about M1/M2 Macs.

I want to remind you that M1 Macs from 2020 are officially unfixable in 2025.

 

[Nermal]

I want to remind you that M1 Macs from 2020 are officially dead and unfixable in 2025.

It's five years from when Apple stops selling that model, so end of life ranges between June 2027 and January 2028 depending on model.

It's still not a great situation though since all the parts are integrated and proprietary and therefore a third-party repairer will be of limited help.

 

[IconDRT]

Apple will release a statement at WWDC saying workflows their customers are interested in have “evolved” and that a traditional workstation design isn’t important for the typical Apple pro. They aren’t taking their time trying to overcome technical hurdles, they are taking their time developing a new reality distortion field narrative that convinces the remaining Apple pros to just adopt a juiced up Studio. For those that aren’t convinced (and didn’t already jump ship in 2013), they will simply move on from Apple, and their defection won’t even amount to a rounding error.

 

[avro707]

It's still not a great situation though since all the parts are integrated and proprietary and therefore a third-party repairer will be of limited help.

A lot of us are suspicious of the 6,1 Mac Pro because of those proprietary GPUs (what happens when they fail), so more proprietary stuff doesn't look good.

On the 7,1 front or even the old 5,1 those are quite easy to support ourselves if we need to.