Like it or not, Apple needs Mac Pro.

[leman]

I'm thinking the main reason for having an Mac Pro (in Intel days) was "the slots".

The primary reason was performance. The Mac Pro used faster CPUs and faster GPUs. Yes, some users mainly cared about expandability (e.g. specialized audio interfaces), but I'd wager that the most users who got the Mac Pro did so because it was considerably faster. That was why we were purchasing Mac Pros for example.

And then it follows, what's the main reason for slots?
Would that not be for graphics cards?

Storage, video capture cards, audio equipment, you name it. Video cards too, obviously, but most folks just used what came with their Mac Pro (you did have more choice as you could use multiple GPUs for more oompf if needed).

I may have this wrong, but are PCI video cards NOT supported on m-series Macs?
And... is this a limitation of the entire design?

If true -- if one can't install 3rd-party video cards in an m-series Mac Pro -- doesn't that severely limit their usability...? And their intended market?

You cannot use third-party GPUs with Apple Silicon, which is by design. To make them usable they would need Metal drivers, and Metal on Apple Silicon makes certain performance and capability guarantees which are not satisfied by third-party hardware. Apple's choice guarantees a streamlined programming model for all the hardware platforms, which obviously comes at a cost.

Another issue is that Apple Silicon SoCs only have very little PCI lanes compared to typical workstation chips. This again is a limitation of the design — there is only that much I/O you can fit on a chip, and on Apple Silicon most of that is taken by the memory controllers. So there is not much bandwidth available to PCI devices (we are talking about 32GB/s or 64GB/s for a Max die, depending on which PCI version is implemented) to begin with. High-performance GPUs are pretty much out of the question here, and if you need very fast networking (e.g. for compute) or ultra-fast storage combined with high-def video capture, you might run into limits too.

 

[throAU]

Macs account for such a small percentage of revenue, one could make the argument that it may not mke financial sense to invest in 10s or 100s of millions on a chip design dedicated to the Mac Pro (or a dedicated desktop chip). What is sad, and disconcerting is that wearables have surpassed the Mac and the iPad is nipping at its heels.

10% of 390b is still a 39 billion dollar division.

It's about 80% of what Goldman Sachs' revenue in 2024.

It's about 2/3 of what Microsoft paid for Activision. PER YEAR.

Even if the mac pro is 1% of mac revenue, its still half a billion dollars a year, give or take.

I think its worth spending a few tens/hundred million investment on to maintain??

It may be a small percentage, but its still a massive amount of money.

 

[maflynn]

It may be a small percentage, but its still a massive amount of money.

Not when we're talking about a 4 trillion dollar company, and not when they're trying to determine how best to allocate resources and R&D. Is it smarter to spend billions on a business unit that occupies 10% of your revenue and has very little growth potential? Or invest in a new initiatives that open up new revenue streams, or even just spending the R&D budget on the product line that occupies the largest percentage, i.e., iPhones.

Its a massive amount of money to us, not to apple. I like word pictures to help visualize the scale of money we're talking about. If you laid a dollar bill end to end, 39 billion dollars will reach the moon nearly 16 times - impressive given that its 238,855 miles away.

Now lets take apple market cap of 4 trillion dollars and see how many times it can reach the moon - over 1,600 trips, That’s roughly 4.4 Moon trips per day for an entire year

My point is when we're dealing with insanely large numbers that these corpations have, our view of money and their view of money is completely different. It doesn't matter that 10% is 39 billion, its the percentage that matters and like I said how apple divvie up its R&D budget

 

[throAU]

Not when we're talking about a 4 trillion dollar company, and not when they're trying to determine how best to allocate resources and R&D. Is it smarter to spend billions on a business unit that occupies 10% of your revenue and has very little growth potential? Or invest in a new initiatives that have open up new revenue streams, or even just spending the R&D budget on the product line that occupies the largest percentage, i.e., iPhones.

With the massive war chest apple has, they can do both. They literally have more money than they know what to do with at the moment, and neglecting a 39 billion dollar division because … reasons? is stupid.

The mac team have their own management structure no doubt, and can handle themselves assuming they get funding, of which there is plenty to go around.

Furthermore:
The Mac currently has MASSIVE growth potential:

Microsoft are totally screwing the pooch with Windows 11, none of the PC vendors have Apple’s purchasing power over OEMs in this current RAM supply crisis and they have a class leading architecture; they’re at least 2-3 years ahead of the PC market and the only manufacturer who has the full stack.

And to add to all this: Microsoft has killed off support for huge swathes of PC hardware. People are in the market for new hardware!

 

[throAU]

The only reason the Mac Pro exists is because Apple have customers that need to buy computers with the I/O stack by the pallet-load. Even despite the price its a bulk-business purchase holder like the iPhone 16e or the base iPad.

In all honesty i suspect one reason for the mac pro existing still is so that existing customers can simply pull out their budget for hardware from 3 years ago, update the quote and present it to management to buy a bunch of new mac pros.

otherwise, the whole question of “what’s this? why are we buying mac studios now?” comes up, when the business has simply been buying mac Pros for 2 decades.

 

[maflynn]

With the massive war chest apple has, they can do both. They literally have more money than they know what to do with at the moment, and neglecting a 39 billion dollar division because … reasons? is stupid.

Evidently Apple's leadership is stupid Report claims that Apple has yet again put the Mac Pro “on the back burner”

The mac team have their own management structure no doubt, and can handle themselves assuming they get funding, of which there is plenty to go around.

I don't know the internal management structure but what is evident is that MAc Pros purportedly only amount to 3% of the overall revenue of the mac business unit. I would fathom a guess that they're focusing more attention on mac products that have a larger contribution to their bottom line.

The Mac currently has MASSIVE growth potential.

Macs may have a rosy outlook, but massive? Sorry I disagree. But we're not talking about Macs in general, we're talking about the Mac Pro and there's plenty of evidence that apple is stepping back from that model.

 

[throAU]

Yeah the mac pro… limited quantity.

Sorry to be clear - i don’t necessarily agree with the premise that Apple “needs” the mac pro per the thread title.

More than hand waving away a 39 billion dollar division as being to small to bother spending say a 100m per year (i.e., 0.25% of the division’s revenue) to keep their hollywood users happy, and keep the reputation amongst high end creatives with a halo product (even if few buy it) is… questionable.

I mean why do Ferrari make things like the F80? Why do VW keep Bugatti around? Why did Porsche build the 959 to sell at a massive loss?

It’s chump change for apple, to keep up appearances.


But yes, i do stand by the assessment that there’s scope for massive growth for the mac right now. Windows users are angry and not all of them are going to get along with Linux. Not by a long shot.

 

[throAU]

Another issue is that Apple Silicon SoCs only have very little PCI lanes compared to typical workstation chips. This again is a limitation of the design — there is only that much I/O you can fit on a chip, and on Apple Silicon most of that is taken by the memory controllers. So there is not much bandwidth available to PCI devices (we are talking about 32GB/s or 64GB/s for a Max die, depending on which PCI version is implemented) to begin with. High-performance GPUs are pretty much out of the question here, and if you need very fast networking (e.g. for compute) or ultra-fast storage combined with high-def video capture, you might run into limits too.

Actually this is a way apple could distinguish the mac pro and actually turn it into something interesting:

Shave off most of the thunderbolt lanes and give it more PCIe. Offer a thunderbolt card for those who want them back.

But this would open the machine up a bit to more useful high speed interfaces like 100GbE (10GbE on a workstation being the best these days = … what?? 10GbE = 1GB/sec … that’s horrifically slow compared to internal storage!), infiniband, fibre channel, etc.

Studio users probably don’t need/want such things. But it would give the Pro a reason to exist - high speed IO available to link them more effectively, and get higher speed networking and SAN access.

 

[leman]

Actually this is a way apple could distinguish the mac pro and actually turn it into something interesting:

Shave off most of the thunderbolt lanes and give it more PCIe. Offer a thunderbolt card for those who want them back.

But this would open the machine up a bit to more useful high speed interfaces like 100GbE (10GbE on a workstation being the best these days = … what?? 10GbE = 1GB/sec … that’s horrifically slow compared to internal storage!), infiniband, fibre channel, etc.

Studio users probably don’t need/want such things. But it would give the Pro a reason to exist - high speed IO available to link them more effectively, and get higher speed networking and SAN access.

Sure, but that would require a fully custom solution just for the Mac Pro. And one still needs to solve the fundamental reason for the lackluster I/O on Apple Silicon — you need space along the die perimeter for connectivity and most of it is taken by the huge memory interface. As I just mentioned in the related thread, a 2.5D/3D SoC design might allow them to split the I/O to a separate die and potentially create more connectivity for the full family.

By the way, Apple has a patent that describes UltraFusion-like arrangement of SoCs, but with a separate I/O die sandwiched in-between, providing additional routing capabilities and PCI lanes. Something like that might work and be cheaper than a custom Mac-Pro-only SoC. Unfortunately, with Extreme being allegedly canceled, I doubt we'll ever see this.

 

[MacCheetah3]

Actually this is a way apple could distinguish the mac pro and actually turn it into something interesting:

Shave off most of the thunderbolt lanes and give it more PCIe. Offer a thunderbolt card for those who want them back.

But this would open the machine up a bit to more useful high speed interfaces like 100GbE (10GbE on a workstation being the best these days = … what?? 10GbE = 1GB/sec … that’s horrifically slow compared to internal storage!), infiniband, fibre channel, etc.

Studio users probably don’t need/want such things. But it would give the Pro a reason to exist - high speed IO available to link them more effectively, and get higher speed networking and SAN access.

Sure, but that would require a fully custom solution just for the Mac Pro. And one still needs to solve the fundamental reason for the lackluster I/O on Apple Silicon — you need space along the die perimeter for connectivity and most of it is taken by the huge memory interface. As I just mentioned in the related thread, a 2.5D/3D SoC design might allow them to split the I/O to a separate die and potentially create more connectivity for the full family.

By the way, Apple has a patent that describes UltraFusion-like arrangement of SoCs, but with a separate I/O die sandwiched in-between, providing additional routing capabilities and PCI lanes. Something like that might work and be cheaper than a custom Mac-Pro-only SoC. Unfortunately, with Extreme being allegedly canceled, I doubt we'll ever see this.

The 2023 Mac Pro has a similar capability (and limitations) as other modern consumer/prosumer PCs. That is, about 20 PCIe lanes available for expansion cards. The other lanes are (by default) allocated to the included/onboard I/O (e.g., networking, USB) — which on “PCs” is often via the “chipset."

Install PCIe cards in your Mac Pro (2023) - Apple Support

Learn about the PCIe slots on your Mac Pro (2023) and how to install additional cards.

support.apple.com

AMD Socket AM5 Chipset for Motherboards

Featuring PCIe® 5.0 for the fastest NVMes, high-speed DDR5 memory, and AMD EXPO™ technology so you can build an advanced PC for gaming and creating.

www.amd.com

You can add a 100Gb NIC:

100Gb FastFrame 4 (model N412) Dual-Port Ethernet SmartNIC (QFP28 incl

QSFP28 module(s) included - Windows and Linux support

shop.atto.com

However, that’s going to consume most of the expansion slot bandwidth.

Same goes for substantial storage expansion:

M.2 4x4 NVMe SSD Silent PCIe Card - Sonnet

Four M.2 NVMe SSD slots on a PCIe 3.0 x16 card. Silent operation. Add your own ultra-fast SSDs for Mac, Windows, and Linux.

www.sonnettech.com

Or, for example, you somewhat increase both:

25/10Gb FastFrame 4 (model N422) Dual-Port Ethernet SmartNIC (SFP28 in

QSFP28 module(s) included - Windows and Linux support

shop.atto.com

M.2 2x4 Low-profile PCIe Card - SONNETTECH

Two M.2 NVMe SSD slots on a PCIe 3.0 x8 card. Silent operation. Add your own ultra-fast SSDs for Mac, Windows, and Linux.

www.sonnettech.com

Or add more USB ports:

Mac Pro (2023/2019) Performance Products

Enhance your 2023/2019 Mac Pro with a Sonnet performance product ranging from add-you-own M.2 NVMe and 2.5" SSDs, to USB 3.2 Gen 2 (10Gbps) and high-speed networking cards.

www.sonnetstore.com

Audio interfaces:

UAD-2 PCIe Accelerator Cards

Get stunning analog sound for your PC or Mac. Run UAD Powered Plug-Ins on compatible Thunderbolt Mac or Windows 10 systems Get a massive DSP boost for powering large professional mixes Access over 100 plug-ins from Ampex,® Lexicon,® Helios,® Studer,® Neve,® Manley,® Capitol Studios,® API, and...

www.uaudio.com

RedNet PCIe Card

For situations in which low-latency performance is critical, such as tracking, the RedNet PCIe card provides the best possible performance for your system.

us.focusrite.com

Which appear to only need X4 width slots — of course, you should be able to install them in the X8 slots.

Most video gear has seemingly gone strictly external, that is, TB/USB4. So, therefore, Mac Pro (i.e., PCIe) is much less relevant, if at all.


On the other hand, if we indeed categorize Mac Pro as an HEDT… Well… it’s nearly undebatable the latest Mac Pro is greatly lackluster in expandability to an AMD Threadripper or Intel Xeon system, which feature 80+ PCIe lanes and eight DIMM slots.

AMD Ryzen Threadripper™ Processors for Desktop Workstations

AMD Ryzen™ Threadripper™ processors deliver battle-tested performance and capability to enable artists, architects, and engineers with the ability to get more done in less time.

www.amd.com

5th Gen Intel® Xeon® Processors – Intel

5th Gen Intel® Xeon® processors deliver more compute and faster memory at the same power envelope as our previous generation.

www.intel.com

In the end, they’re tools. And you select them based on what you need, can do without/compromise, want, and dislike.

 

[throAU]

You can add a 100Gb NIC:

100Gb FastFrame 4 (model N412) Dual-Port Ethernet SmartNIC (QFP28 incl

QSFP28 module(s) included - Windows and Linux support

shop.atto.com

However, that’s going to consume most of the expansion slot bandwidth.

Yeah, figured you probably could, but imho the pro mac would be better off just trimming the onboard thunderbolt ports down to say... 2-4 and using the lanes for more PCIe and/or replacing the onboard 10GbE with 25/40 or 100GbE options via onboard QSFP/SFP.

10GbE on a pro workstation these days is pretty low end. For the sort of work i'd be aiming the Mac Pro at i'd be wanting 100GbE to a storage array and if you're already low on ports at that point then the thing is kinda bandwidth starved.

And who is seriously going to be using a heap of thunderbolt, by choice on a big box mac with PCIe slots? If they do need it, plug some more in via PCIe?

 

[MacCheetah3]

Yeah, figured you probably could, but imho the pro mac would be better off just trimming the onboard thunderbolt ports down to say... 2-4 and using the lanes for more PCIe and/or replacing the onboard 10GbE with 25/40 or 100GbE options via onboard QSFP/SFP.

10GbE on a pro workstation these days is pretty low end. For the sort of work i'd be aiming the Mac Pro at i'd be wanting 100GbE to a storage array and if you're already low on ports at that point then the thing is kinda bandwidth starved.

And who is seriously going to be using a heap of thunderbolt, by choice on a big box mac with PCIe slots? If they do need it, plug some more in via PCIe?

I don’t have any disagreements to your thinking. However:

Each built-in Thunderbolt port in Mac Pro is managed by its own controller integrated in the M2 Ultra chip and doesn't share bandwidth with the PCIe slots.

In other words, the idea of exchanging Thunderbolt ports for allocatable PCIe lanes seemingly falls under:

Macs account for such a small percentage of revenue, one could make the argument that it may not mke financial sense to invest in 10s or 100s of millions on a chip design dedicated to the Mac Pro (or a dedicated desktop chip).

From what I understand, the eight Thunderbolt ports (6 rear, 2 top-front) on the Mac Pro exist because it’s solely configurable with the M- Ultra SOC, which — as you probably already know — is two interconnected (“UltraFusion”-ed) M- Max SOCs. Each M- Max (and M- Pro as well as some base M-) SOC includes controllers that provide four Thunderbolt ports — I’ve read it’s one Thunderbolt controller per two ports.

Therefore, coming full circle, presumably, unless Apple designs a Mac Pro-specific SOC or adds an I/O subprocessor/die to their UltraFusion packaging, your idea/preference would result in every non-Ultra Mac to be limited to two Thunderbolt ports with excess PCIe lanes.

How simple or not would it be to expand the capability of the PCIe host controller without encroaching on Thunderbolt?
🤷‍♂️

 

[maflynn]

Therefore, coming full circle, presumably, unless Apple designs a Mac Pro-specific SOC or adds an I/O subprocessor/die to their UltraFusion packaging, your idea/preference would result in every non-Ultra Mac to be limited to two Thunderbolt ports with excess PCIe lanes.

It seems fairly clear that apple has not been interested doing that since the introduction of 2023 mac pro, in fact we've not seen any updates to that mac, while other computer models were updated to the M3, M4 and now the M5 is rolling out.

There was supposedly a Mac Pro specific SOC but its never seen the light of day - the rumored M series Extreme cpu.

 

[Unregistered 4U]

Call me crazy… But… I’m thinking they don’t.

 

[1BadManVan]

Call me crazy… But… I’m thinking they don’t.

lol, yea this post didn't age well unfortunately

 

[deconstruct60]

lol, yea this post didn't age well unfortunately

This thread hadn't aged well when this particular thread started.

The notion that a highly specialized and proprietary AI server chip was going to also result in a Mac Pro optimized SoC was flawed from the previous conversations , before it got respawned in this thread.

"They need it for training". ... after
https://www.macrumors.com/2025/11/05/apple-siri-google-gemini-partnership/

Gemini is already baseline trained on non Apple Silicon. They don't need Apple Silcion for more training. (Google pays billions to Apple anyway. Billions in training compute time isn't much different than collecting dollars. )

the more heavy the derviative of Gemini the more might not even need their own AI server chip. There are lots of hyperscalars rolling their own AI server niche chips , so Apple may end up keeping one for themselves. But the more specialized it is , the more likely stuff like Display engines would get dumped for more pressing "AI datacenter needs" . Hence, don't end up with a single user, GUI OS chip. It is a multiple user , no-GUI OS chip.

 

[1BadManVan]

This thread hadn't aged well when this particular thread started.

The notion that a highly specialized and proprietary AI server chip was going to also result in a Mac Pro optimized SoC was flawed from the previous conversations , before it got respawned in this thread.

"They need it for training". ... after
https://www.macrumors.com/2025/11/05/apple-siri-google-gemini-partnership/

Gemini is already baseline trained on non Apple Silicon. They don't need Apple Silcion for more training. (Google pays billions to Apple anyway. Billions in training compute time isn't much different than collecting dollars. )

the more heavy the derviative of Gemini the more might not even need their own AI server chip. There are lots of hyperscalars rolling their own AI server niche chips , so Apple may end up keeping one for themselves. But the more specialized it is , the more likely stuff like Display engines would get dumped for more pressing "AI datacenter needs" . Hence, don't end up with a single user, GUI OS chip. It is a multiple user , no-GUI OS chip.

to clarify, my comment was in agreement with you about the thread, not your particular post lol

 

[MayaUser]

Mac Pro is officially dead, along side the idea of getting a M Extreme or whatsoever
This thread can be closed now

 

[deconstruct60]

Mac Pro is officially dead, along side the idea of getting a M Extreme or whatsoever
This thread can be closed now

The M Extreme never made it to the MP 2023. ... that was dead a long while back. Too few customers and too expensive to make. Was dead in the water even before the Mac Pro showed.

 

[maflynn]

Despite that Mac Pro is being on the back burner, Apple will need to develop high-end and/or workstation grade Apple Silicon chips for their own future. I would skip the performance and specs of Macs cause it's gonna be a long conversation despite the fact that Mac's max performance is poor but the most important fact is, they need workstation grade chips to make their own servers for AI.

Looks like the disposition of the Mac Pro has been settled and Apple has now decided its not worth keeping this in their line up.

I can't say that I'm surprised, I've largely said as much. With the advent of Apple Silicon, a tower form factor with lots of bays room for expansion cards, cooling, and ram expansion is not needed.

 

[IvyKing]

The M Extreme never made it to the MP 2023. ... that was dead a long while back. Too few customers and too expensive to make. Was dead in the water even before the Mac Pro showed.

The linked Mac Studio's seem to be Apple's answer to the M extreme.

 

[deconstruct60]

The linked Mac Studio's seem to be Apple's answer to the M extreme.

A four node each with a point-to-point TBv5 proprietary RDMA connnection will run certain workloads faster. GPU graphics out probably isn't one of them. Some folks want the Nvidia x080/x090 killer Extreme for 3D graphics. that isn't going to work. Even chiplet GPU in a single package are quirky. The latency and bandwidth limitation os TBv5 isn't really viable at high refresh, max colors , at max resolution.

But yeah. If clustering 4 on a desktop that will pratical for Studio node. For Mac Pro nodes it is not because the system volume is much bigger.

That said I don't really think it is "Apple's answer" as much as Apple changing the market target to being a substantively different set of users. New workloads and new users that want those workloads as opposed to taking long term legacy upper end Mac Pro workloads. Somewhat like a pro sports team swapping players during free agency period. Team still numbers the same amount of players before the shuffle but it is a different team.

 

[Ursadorable]

Just out of curiosity, what PCIe cards actually work and are actively supported on the Apple Silicon Mac Pro?

 

[maflynn]

Just out of curiosity, what PCIe cards actually work and are actively supported on the Apple Silicon Mac Pro?

Networking, and storage cards, Video I/O cards (not GPU) like the stuff from BlackMagic, and Audio cards

All of which are available via thunderbolt, so even then there's viable alternatives.

 

[Unregistered 4U]

The M Extreme never made it to the MP 2023. ... that was dead a long while back. Too few customers and too expensive to make. Was dead in the water even before the Mac Pro showed.

M Extreme was rumored by folks that had no information BUT were aware of how other chipmakers worked and just assumed Apple would want a system just to post numbers for.

I should place a bet on Polymarket as to “Which year will Intel finally release a lineup where every chip they produce is more performant than the M1?”