MP All Models The Mac Pro is Dead, This is the End?

[MoonCakeTropics]

To an extent - but a few people still need PCIe slots for specialist I/O, audio/visual and networking cards or to add large quantities of internal storage. That's the only reason for buying a Mac Pro post-Apple Silicon. The PCIe bandwidth/number of lanes/number of slots supplied by the Mac Pro still greatly exceeds what you can get from a Thunderbolt 3 to PCIe enclosure. Thunderbolt 5 might help a bit, there esp. for PCIe 4 capable cards once suitable enclosures appear.



If you submit to Apple's "integrated GPUs rule" theory then 16 lanes is a fairly generous provision for specialist A/V cards, network cards and SSDs - which is all you can plug in to the Mac Pro.

Your typical x86 system will need at least one discrete GPU which will eat 16 lanes. High-end ThreadRipper Pro machines (...and even the old 2019 Mac Pro) will often have multiple GPUs each sitting on 16 lanes which is the main reason that Threadripper, Xeon-W etc. need so many lanes.

I disagree with this in part. For me, having fast u.2 storage and fast networking make the studio a non starter. PCIe5 based u.2 drives are now between 60-120tb of storage. And are super fast. Best bang for buck at fast speeds by far. The 16tb offering in the studio is likely stuck at PCIe 4 speeds, and for that price you can likely get two 30tb U.2 drives that would trounce apple's offerings both in speed and capacity.

Anyway, the controller cards like from highpoint are 16 lane just for the SSDs. Then if you want state of the art fiber networking, which is now up to 800gbe, youre going to need 16 lanes as well. And, this is not crazy, if you want two cards so you can really get crazy bandwidth on the SSDs, you'd like another 16 lanes.

So while I agree with you some of the still good use cases for a Mac Pro with PCIe slots are storage and networking, even if you need zero GPU prowess, the number of lanes in the 2023 are just on the anemic side. Sure, it's better than nothing, and even if you share some bandwidth, on average it will still be much better than what a studio could do via TB5, but it would be nice to increase the number of lanes a bit.

 

[MoonCakeTropics]

Every time Apple doesn’t update a product for a while, everyone starts claiming it’s dead.

For years, as Apple was planning the Apple Silicon transition, there was a chorus of people crying that Apple no longer cared about the Mac and was abandoning it in favor of iOS. During this time, Apple stated over and over than they were committed to the Mac and had no plans to merge it with iOS, but the skeptics continued to claim the Mac was dead and that a transition to ARM was impossible. Fast forward a few years, and now we’re in a golden age of the Mac.

The point is, Apple doesn’t do updates for updates’ sake. They wait until a true successor is ready, while many people can’t see past the short term. Is the Mac Pro a niche product for a small subset of users? Yes. Does that mean it won’t get updated as often as more mainstream products? Yes. Does that mean it will never get updated? Probably not. Does that mean that Apple doesn’t care about making high performance workstations? Definitely not. Apple is just waiting until there is a material update to make and will continue to push the high end of performance. The complaints that Apple is abandoning “pro” users in favor of mainstream products are as old as time and always prove to be wrong.

TDLR; relax, ye of little faith.

Tell us more about: MPX slots, mac cube, the imac pro, airport extreme, back2mac, 12inc mack book, 27" imac, time capsule, apple printers, Xraid, ipods, newtons, apple lisa, touch bar, 11" macbook, etc etc

 

[MoonCakeTropics]

Shocking lack of vision presented here.

Yes, the M2 Mac Pro was an embarrassment. And Apple should have refreshed it once more with updated X86 guts in 2022 so they could run it another 3 years, just in time for its grand makeover again for Apple Silicon when it was truly ready. But I suspect they really felt compelled to stick to their self-imposed roadmap.

It's still possible that they let it die, I suppose. I really hop that's not the case though. I'm hoping we see something roughly like the following:

M5 generation, as mentioned above, ends up being the architecture that really kicks things into high gear, and allows them to take on proper workstation and server-grade workloads. To really drive this home, they will invert the rollout from mobile-first to desktop-first. The most powerful machines will get it first, and then it will roll out to the lower powered machine over time as costs go down.

They will start with an M5 Pro, M5 Max and M5 Ultra Mac Pro at WWDC. Maybe tops out close to 100 CPU cores, and some multiple of that in GPU cores. 2tb of RAM, TB5, etc. The real killer app, however, would be using the "MPX" form factor for expansion modules. Examples might be - just another M5 module, modules devoted to GPU cores, modules devoted to AI/tensor cores, maybe other specialized functions like audio in/out, video capture, etc. Maybe 3rd parties get involved again.

This would almost certainly allow Apple to decisively take back the crown for "most powerful desktop machine on earth" and allow them to compete with Nvidia, Intel and AMD for the most demanding workloads. Something like this would truly change the game and re-invent what a personal computer is capable of, again.

Not sure if anything like this is on the table but if they have any drive and motivation left over there, they will hopefully try something like this. I'd certainly like to see this.

We already know that M5 will form the basis for their in-house servers, to me that makes it the perfect candidate for a product like this. You could end up with a desktop machine that, when fully spec'd out, could go toe-to-toe with GB200. Hundreds of CPU cores, many hundreds of GPU cores, terabytes of memory. The appetite will absolutely exist for machines like this, especially if they are willing to help on the software side.

I love your optimism and hope you are right. But what makes you think the M5 is 'that chip'? If the rumors are to be believed, the M5 is another iterative chip at 3nm. The M6 is rumored to be the big jump to 2nm. Not sure you need a fab jump to do any of what you say, but it certainly would bring more 'breathing room' to their designs. I hope youre right, but I havent even heard any rumors anywhere near what youre presenting. If you have, I'd super appreciate a link.

 

[mattspace]

Do you have data to support your claim that most people buying an iPad don't want iPadOS?

Do you have data to support the notion that anyone wants iPadOS on their iPad?

 

[mattspace]

Lotta people who've never participated in the Mac Pro forums, and who don't appear to own Mac Pros, showing up with big opinions all of a sudden.

 

[mattspace]

Really? "Most people"?

What's your basis for this assertion? I can honestly say I've never met anybody who has said "I want an iPad but I really don't want to have to run iPadOS on it".

Maybe it's more concentrated amongst the people who make things. All I hear from folks with iPads, especially iPad Pros is how much they're like trying to run waist deep in water.

But iPads are off topic, realistically and the only relevence they had to the conversation in the first place, IIRC was that that Macs are effectively just iPads with different specs now (so why would you expect a Mac to support commercial alternative operating systems), which are allowed to boot into an operating system that is increasingly just a skin for an iOS variant.

Despite what Apple says about iOS and macOS not merging, that is what is happening, just not at the UI level, so as to maintain the need for two devices to do what Windows does just as well with one.

Those who actually own Mac Pros would be well aware of our recent experience of an entire OS cycle with no reliable PCI storage, which was due to a change in the way the OS handles non-apple storage devices, likely because an older macOS specific part of the OS was replaced by an iOS derivative that was only ever designed to work with cabled external drives.

 

[MoonCakeTropics]

"Ass backwards" would be to fabricate the most complicated chip first on a new process.

The reason we have an M3 Ultra and not an M4 Ultra today is because the M3 fabrication process has been ironed out and wafer yields and capacity are now high enough you can manufacture them for a similar cost structure to what M2 Ultra cost.

And we know these M3 Max die are not the same as those in a MacBook Pro since they have TB5 and new memory controllers that can access four times as much memory. I also would not be surprised if Apple announces M3 Ultra supports PCIe 5 compared to the PCIe 4 of M2.

Yes, in a world where convenience for engineers trumps customer needs, that would be accurate. If that is the actual case, it tells you more about apple's mismanagement than their marketing prowess in serving their customer base.

 

[avro707]

Lotta people who've never participated in the Mac Pro forums, and who don't appear to own Mac Pros

That was my thought as well. Hence why I asked early in the topic to see who actually has 7,1 Mac Pros.

 

[impulse462]

Do you have data to support the notion that anyone wants iPadOS on their iPad?

Well tbh the proof is that they bought an ipad, but in general i agree with your previous post that apple clearly locks these down to "force" people to buy both. i personally think macos would work pretty well on an ipad just with a few tweaks. i understand some dont agree with me which is fine, but a lot seems to be parroting apple marketing. apple will eventually come out with some macos derivative on the ipad and everyone will suddenly clamor and say its revolutionary. this is a well documented effect.

 

[mattspace]

Well tbh the proof is that they bought an ipad

Yeah, but without an actual OS choice there's no evidence that users want the specific OS. Theres no more reason to think people specifically want iPadOS than people want Android Auto, because that's what powers the satnav in the car they buy... except of course when they can use Carplay and we can see that when given a choice they don't use the default hardware OS.

but in general i agree with your previous post that apple clearly locks these down to "force" people to buy both. i personally think macos would work pretty well on an ipad just with a few tweaks. i understand some dont agree with me which is fine, but a lot seems to be parroting apple marketing. apple will eventually come out with some macos derivative on the ipad and everyone will suddenly clamor and say its revolutionary. this is a well documented effect.

Yeah, there's a real this is bad and everyone doing it is wrong until Apple does it, and suddenly the underlying ideas are all validated because Apple did them, that infects professional (i.e. Apple "journalists" and purchased commentators) and amateur Apple fanboys alike.

 

[CWallace]

Well Gurman did say in 2024 that Apple was considering releasing the Mac Studio with the codename for the M4 Max and "an M3 era chip", both of which which did come true. And as part of that report, he did say the Mac Pro would get a custom version of the M4 with the codename "Hidra" so maybe we will be surprised come WWDC.

 

[Regulus67]

Well tbh the proof is that they bought an ipad....

I have two iPad Pro M1. And yes, I would not have done that if it came with macOS. And I never had a macbook of any kind.

 

[Allen_Wentz]

I'd like a Mac with lots of RAM, to play to my current interests in LLMs and other AI. Such a thing does not look to be coming, though. And yes, the 2019 Mac Pro can take a lot of RAM but it's pretty slow compared to Apple Silicon.

You can get a Mac now with lots of RAM. Apple retrofitted 500 GB RAM to the Studio M3 Ultra. The fact that they did that, along with the large jump in available RAM even for laptops suggests to me that Apple is likely to do something more about making RAM available as they move to M5 chips.

What would be cool is if along with making more RAM available they dramatically reduced the cost of doing so.

 

[Allen_Wentz]

Macs are effectively just iPads with different specs now

I way disagree. Macs came first by decades and are still the side with real files management. So no, Macs are not effectively just iPads. I do understand why children who grew up teething with iPads might think so, and be particularly comfortable with iPadOS.

 

[Regulus67]

I'd like a Mac with lots of RAM, to play to my current interests in LLMs and other AI. Such a thing does not look to be coming, though. And yes, the 2019 Mac Pro can take a lot of RAM but it's pretty slow compared to Apple Silicon.

How slow do you recon it is in comparison?

I am reading about running DeepSeek R1 671 billion parameter model locally.
hardware-requirements-for-deepseek-r1
running-deepseek-r1-671b-locally

The 2019 Mac Pro looks useful.
It can fit 1.5 TB RAM, two dual W6800X (128GB VRAM), and fast NVMe SSDs.
I do not mind if it slow, as long as it is capable of training on my machine. And I will be using RAG (Retrieval-Augmented Generation).

______________________________________
High-End Multi-GPU Server:

• • CPU: Dual Intel Xeon or AMD EPYC
  • GPU: 8x NVIDIA RTX 3090 (NVLinked in pairs if possible)
  • RAM: 512GB – 1TB DDR5
  • Storage: Multiple fast NVMe SSDs in RAID0 (optional, but may not provide significant gains)
  • Inference Engine: vLLM or custom implementation with tensor parallelism
• This configuration maximizes performance with multiple GPUs and ample memory, enabling faster inference with larger models and context lengths.

 

[mattspace]

I way disagree. Macs came first by decades and are still the side with real files management. So no, Macs are not effectively just iPads. I do understand why children who grew up teething with iPads might think so, and be particularly comfortable with iPadOS.

Seriously, have you not read the entire thread? I've said it's about the hardware. Mac hardware is just expanded (and sometimes identical) iPad hardware, running a different operating system.

 

[neuropsychguy]

Do you have data to support the notion that anyone wants iPadOS on their iPad?

With all due respect, that’s a terrible reply. First, I didn't make any claims about people wanting iPadOS on their iPads. I simply asked for evidence to support the claim that most people do not. So your follow-up question is a red herring.

When a claim is made that a majority of iPad owners don’t want iPadOS on their devices, the burden of proof is on the person making the claim to back it up with data.

In any case, the reasonable assumption is that people buying an iPad, iPhone, Apple Watch, Mac, etc. want the operating system on the device. Claiming that "most people" who buy those (even just an iPad) would prefer a different operating system is a claim that requires evidence.

A snarky retort like yours does nothing to support the claim or be part of a reasonable discussion.

 

[mattspace]

When a claim is made that a majority of iPad owners don’t want iPadOS on their devices, the burden of proof is on the person making the claim to back it up with data.

I would say the fact the entire history of the iPad is progression from a direct install of the same operating system and UI the iPhone has, to implementing more, and more of macOS features (filesystem, mouse, keyboard, extra displays etc.) shows that Apple is pretty clear that what peope want from their iPad, is for it to be more like a mac, than it's like an iPhone, or indeed like the original iPad.

The iPad today more closely resembles the paradigm of the Surface Pro that existed when the iPad launched, than it does the original iPad.

In any case, the reasonable assumption is that people buying an iPad, iPhone, Apple Watch, Mac, etc. want the operating system on the device. Claiming that "most people" who buy those (even just an iPad) would prefer a different operating system is a claim that requires evidence.

No, it isn't. That's an utterly untenable claim, because there is no alternative to measure it against. On an Intel Mac you can claim that most people want macOS over Windows, because Windows was an equally supported alternative.

 

[MoonCakeTropics]

I disagree with this in part. For me, having fast u.2 storage and fast networking make the studio a non starter. PCIe5 based u.2 drives are now between 60-120tb of storage. And are super fast. Best bang for buck at fast speeds by far. The 16tb offering in the studio is likely stuck at PCIe 4 speeds, and for that price you can likely get two 30tb U.2 drives that would trounce apple's offerings both in speed and capacity.

Anyway, the controller cards like from highpoint are 16 lane just for the SSDs. Then if you want state of the art fiber networking, which is now up to 800gbe, youre going to need 16 lanes as well. And, this is not crazy, if you want two cards so you can really get crazy bandwidth on the SSDs, you'd like another 16 lanes.

So while I agree with you some of the still good use cases for a Mac Pro with PCIe slots are storage and networking, even if you need zero GPU prowess, the number of lanes in the 2023 are just on the anemic side. Sure, it's better than nothing, and even if you share some bandwidth, on average it will still be much better than what a studio could do via TB5, but it would be nice to increase the number of lanes a bit.

Just to follow up on the lane question, apparently PCIe 6 already working and wow:

Micron shows off world's fastest PCIe 6.0 SSD, hitting 27 GB/s speeds — Astera Labs PCIe 6.0 switch enables impressive sequential reads

The next-gen of networking and storage is hitting the trade shows

www.tomshardware.com

Micron's PCIe 6 drive surpass expectations and measure read speeds of 27.14 GB/s. For context, the fastest PCIe 5.0 SSD that we've tested, the Crucial T705, hits maximum read speeds of 14.5 GB/s, maxing out the PCIe 5.0 x4 connection...
...
The drive was pushed to this point thanks to Astera's Scorpio P-Series Fabric Switch, an industry-first network switch at PCIe 6 that connects up to 64 PCIe 6.0 lanes.
...
Today's PCIe 5.0 maxes out at 128 GB/s of bidirectional speeds on an x16 bus. PCIe 6.x doubles this, reaching 256 GB/s over an x16 connection.

 

[Ethosik]

Pouring many millions into a niche product like the Mac Pro and its chip each year or so isn't necessarily great management.

Well it’s their fault for making it so niche. Seriously how many times have people begged for just a basic i7 or i9 based desktop with PCIe? But no they just locked it to an insanely priced Xeon, started with laughable 256GB SSD at launch in 2019 and I believe even 8GB of RAM. And it still costed $6,000. This is why I went for a fully maxed i9 iMac instead.

 

[Ethosik]

Lotta people who've never participated in the Mac Pro forums, and who don't appear to own Mac Pros, showing up with big opinions all of a sudden.

I had Mac Pros until the 2013 one. Then I went back to windows. I still had some Mac’s as I preferred Final Cut Pro X. Yes even back then. Hatred for Windows began. Then the 2019 Mac Pro came out and I saw the specs and price. Said nope. Built a better PC for half the price. Hatred for Windows got worse. Apple silicon came out. Immediately got everything replaced as soon as they released them. Mac mini M1 first, then M1 Max laptops. Three of them. Then M1 Ultra Mac Studio.

Now I have 8 computers in my workflow and lab. 6 Macs and two PCs. One of them is demoted to just gaming now. The other PC barely gets any use. I just hate windows that much.

So long story short, while I don’t have a Mac Pro I always keep an eye on it. I had them before the 2013 trash can. Ever since then Apple has just been overly frustrating me with their pro line. I thought things would change with the Apple Silicon but nope.

 

[throAU]

The ultra isn’t about single core performance. It’s the same as the regular so if you need that, you wouldn’t buy the ultra. It would be nice if they can make an ultra with considerably faster single core performance.

Physics exists. Smaller processors with less cores will always be as good or better at single core within the same generation as they can clock higher and be cooled better.

Larger processors of the same generation come out after smaller ones as they are riskier and more expensive and difficult to manufacture. Companies shake the bugs out with mobile and smaller core count chips first.

Intel does this. AMD does this. Apple does this. Because of production and design reasons.

 

[russell_314]

Physics exists. Smaller processors with less cores will always be as good or better at single core within the same generation as they can clock higher and be cooled better.

Larger processors of the same generation come out after smaller ones as they are riskier and more expensive and difficult to manufacture. Companies shake the bugs out with mobile and smaller core count chips first.

Intel does this. AMD does this. Apple does this. Because of production and design reasons.

I’ve seen this with older Mac Pros if you got the top end processor with the most cores, the clock speed was slower. Don’t ask specifically which generation it was because I was window shopping at the time and thought it was interesting that the most expensive processor wasn’t the fastest.

 

[Allen_Wentz]

Yes, in a world where convenience for engineers trumps customer needs, that would be accurate. If that is the actual case, it tells you more about apple's mismanagement than their marketing prowess in serving their customer base.

Yet Apple's business operation is essentially a case study in using "marketing prowess in serving their customer base." Apple did not get to $4T by mismanagement !

 

[Allen_Wentz]

I’ve seen this with older Mac Pros if you got the top end processor with the most cores, the clock speed was slower. Don’t ask specifically which generation it was because I was window shopping at the time and thought it was interesting that the most expensive processor wasn’t the fastest.

Exactly. Computers above basic need to be configured to suit intended workflows. The new M3 Ultra with 500 GB RAM available is a good example. If one needs 300 GB RAM or whatever for a workflow then an M4 Max with 128 GB RAM is a weaker solution.