Apple Executive Discusses New Mac Pro's Lack of Graphics Card Support

[crsh1976]

Seems like a weird decision, at least it can still grate thicc mozzarella shreds.

I think the weirder decision is to keep the Mac Pro around at all, knowing the architecture route Apple has taken for its own chips does not work for modular setups such as scalable/expandable workstations.

This Mac Pro, with its 192 GB RAM limit and its open, near useless PCIe slots, isn't going to fare well.

 

[CarAnalogy]

How do you not plan for the most common device used in PCIe slots? WTF is wrong with Apple?

This is just corporate speak. He knows damn well what he did. I just wish I could hear the actual discussions they had about this. We'll never know the real reason they did this (not for many years anyway.) The only explanation that makes sense to me is that some of the senior management wants the Mac Pro to go away but they had already promised it to the public, so they did this.

When they don't sell any because this machine isn't made for anyone, they'll use those sales numbers to discontinue it. Even amongst people who bought the last several, I've yet to hear any of them say they want this one. There is no reason to get it when the Mac Studio is the exact same thing. Nobody needs $3000 of PCI slots that can't take graphics cards.

 

[daveedjackson]

A producer (you can check his video on youtube) said that he was using between 300 and 600 in RAM, so for him 192 is a no go.

I also do motion anf VFX and can often use a fair chunk of my 256gb of ram. So there defo is a need/use for more RAM.

 

[Hopefully Smarter]

the ability to not answer a question is the hallmark of an apple executive

 

[threesixty360]

Right, but this only applies on a small scale for regular users. If we move towards things like Machine Learning, AI - for example training models and so on or something like 3D modeling and rendering, there's a massive difference between 190GB vs. 1,5TB. And this is a machine aimed at professionals, the M2 family is great including the Ultra model however for specific use cases there are severe limitations, GPU and RAM being the obvious ones.

is there there's a massive difference between 190GB vs. 1,5TB on a SOC that has high speed memory access in it?

The OS gives the software the memory. If the OS can work in such a way that you can access 1TB of ram by doing some memory swapping under the hood then it's all good right? I mean, your OS doesn't just shut down when you run out a ram does it? It manages your memory.

The problem is everyone writes software using their own custom libraries that they compile on to Windows based architecture thats been roughly the same for 30 or 40yrs. No one has time to optimise anything anymore. So the general public think that Windows is the benchmark and the only way anything can be done.

I bet if these tools were written with Apple Silicon and ARM in mind from the beginning they would have the same performance on far less resources. It's the same issue with security, windows convinced everyone that self propagating viruses were normal for all computers. They were not. Windows OS was designed in a way that made it easy to make them.

Obviously, the big ML tools are never going to write specifically for a Mac because of market share so it will be hard to prove Apple silicon performance. But I just wish sometimes that Windows didn't take over the world because it's made computing so stagnant in some ways as we always develop for the lowest common denominator.

 

[CWallace]

Either way, without GPU’s industrial/medical/other can’t do what they need, therefore Apple is no longer in “that” area of “Pro” business.

But was Apple ever really "in" the industrial/medical/other market when they only offered one single model? This forum was not lacking of people who stated they were in those markets crapping on the Mac Pro because it could not support one nVidia card, much less up to four; that it "only" supported one Xeon CPU and that CPU "only" supported up to 28 cores; that it "only" supported 1.5TB of RAM; etc.

If Apple was really serious about these markets, they would have competed with the entire range of workstation products from HPE, Dell, Lenovo and others with configurations that offered multiple Xeon Gold and Platinum CPUs and multiple terrabytes of RAM and massive power supplies capable of supporting four workstation-class video cards.

A/V pros and that’s it. They just refuse to say it. They are segmenting their pro use. Time to move move

A/V pros seem to be Apple's traditional core market for the Mac Pro and that is why they tried things like the 2013 Mac Pro (which more easily fit into a recording studio or editing bay) and focused more on quiet operation than bleeding-edge performance because these machines were often not locked away in an air-conditioned data closet but instead sat on the desk or directly under it.

And this is why they launched the Mac Studio. I mean the name alone describes the market the model is aimed at: A/V pros and graphic artists. It is effectively the 2013 Mac Pro with a decade's worth of technical improvements. But as Apple learned with the 2013 model, there is a small portion of their core market that needs internal PCIe expansion so rather than just screw them over yet again (only four years after finally apologizing to them for 2013), they released a "Mac Studio with internal PCIe expansion".

Higher end professionals dealing in specific fields who need to work locally without proxies will not be super excited about this as both RAM limitations and GPU limitations are quite crippling. A nice machine for various artists however if you need the RAM M2 Ultra cannot do anything to compensate for it.

See the above. The 2019 Mac Pro was a general purpose workstations because that is all Intel offered in terms of CPUs and motherboards and Apple had to take what they offered. And even there, Apple tweaked it as much as they could to meet the needs of their core market, which was mostly not the market that bought general purpose workstations because that market wanted more variety in configurations and capability than a single model range could offer and who could easily get that variety from companies like HPE, Dell and Lenovo.

It is essentially a Mac Studio with a $3,000 case.

A $3000 case that includes PCIe and SATA expansion with improved cooling and more-than-sufficient power overhead to support that expansion. And yes, that's a lot of money for those options (especially since Apple is leveraging a fair bit of existing materials), but to be sure they did not pick that price in a vacuum - they know the market who will buy these machines will pay those prices.

Apple are really losing sight of a lot of markets.

Markets they were never really in if you looked at them objectively.

The iPhone and the greed of Cook are really killing Apples computing division.

And yet Mac sales and revenue have consistently been on an upward trajectory over the past decade...

They need to stop telling customers what Apple thinks they need and instead listed to what the customers need.

They do listen to their customers, but they choose not to deliver a product for every single one of them, instead addressing the needs of broad ranges of customers.

 

[Longplays]

And how many of those workstation purchases are for boring cheap boxes running office for big corporations? What proportion of those require Mac Pro like specs? I bet it’s a small number…

The table was provided to give Mac Pro users insisting on 2019 features missing on the 2023 an idea how large the worldwide PC workstation market is relative to the PC market as a whole that is 7.6562 million workstations out of 292.3 million PCs.

For those wondering it is a rounded up 2.62%.

When applied to the 28.6 million Macs shipped in 2022, Apple's pro desktop units shipped for both Mac Studio & Mac Pro would be a rounded up approx 75,000 for 2022.

Of which 80% of that are Mac Studio (60,000) & 20% are Mac Pro (15,000).

Let us assume the vocal users insisting on 2019 features are just 20% of that 15,000 Mac Pros.

3,000 conservatively?

Absurd numbers to put in any additional Apple Silicon R&D for Mac Pro specialized features.

But I saw elsewhere on MR with claims that MPX Modules for the 2019 sold 80,000 units. Let us assume that is for a period 3.5 years. This covers Dec 2019 to Jun 2023.

So that is what... nearly 22,900/year? Around 100,000/year Mac Pros in total?

So let us make a higher assumption for the Apple pro desktops to 100,000-200,000 annually.

Apple will make added Apple Silicon R&D spend that does not directly improve revenue for that few users?

They do not profit from sales of 3rd party RAM & SSD as they are standard parts. They may sell rounded up 23,000/year MPX Modules for AMD GPUs that are ballpark performance of M2 Ultra GPU cores.

AMD likely moves more Threadrippers or Epyc chops on PC workstations than Apple does Mac Pro & Mac Studio combined. AMD's a very small player vs Intel.

Lower economies of scale necessitates more expensive goods. Apple needed to prioritize their key use cases that actually buy the Mac Pro.

Does Apple care about Mac Pro users who want 2019 features returned to the 2023? It will be a net LOSS to Apple.

AMD/Intel/Nvidia are better situated to service them. It is a net gain for Apple to not sell to the vocal minority.

 

[Alan Wynn]

I also do motion anf VFX and can often use a fair chunk of my 256gb of ram. So there defo is a need/use for more RAM.

Just to clarify:

1. Is that on a 2019 Mac Pro?
2. If so, did you purchase that RAM from Apple?

 

[Omega Mac]

What if Apple starts releasing their own PCIe based cards for extra GPU or M-Chips in your mac-pro, if that is possible. Since they're making all their own silicon why not produce upgrade cards where you can beef up your mac-pro to something insane using more Apple Silicon?

Apologies if GPU's don't use PCIe I'm not up on the latest but you get the idea.

 

[Chazz12]

I just don't think Apple is going to need dedicated GPUs moving forward. My Mac Studio does all of my games (that do work on Apple Silicon like Eve and No Man's Sky, etc...) work on at least 1440 at the highest settings. I usually turn everything to the highest I can when I start a game. lol. Most of them run at 4k or native resolution on my Studio Display.

 

[TheIguana]

The GPU is one confusing part of this calculus, but I think the bigger loss is the inability to upgrade the RAM. This is a core tenant of truly professional computers as computing requirements scale up. I also think it runs completely counter to the “environmental“ goals of Apple, as it makes it much more likely for a computer to end up in a dead-end e-waste situation because it can‘t be reconfigured to a new use-case.

 

[ectoplasmosis]

Even amongst people who bought the last several, I've yet to hear any of them say they want this one. There is no reason to get it when the Mac Studio is the exact same thing. Nobody needs $3000 of PCI slots that can't take graphics cards.

We need exactly this, and have ordered fair quantity of the new Mac Pros to replace our current Mac Studios with external PCIe chassis.

We use multiple SDI/SMPTE2110 video capture/output cards, 25Gb/100Gb network cards and NVMe HBA cards, all of which are crippled to varying degrees by the 4-lane Thunderbolt bus, and have been waiting for exactly what the new Mac Pro provides. We're hardly a fringe case either; from speaking with friends in the industry, everyone using MacOS is buying them up. The Mac Studio is fantastic, but PCIe support is crippled; the Mac Pro fixes this.

You're not the target market, and fail to imagine a use case. Real-world customers are buying the product while you complain on the internet.

 

[chrono1081]

I understand the differences between ARM and x64 architectures quite well but you are saying is simply not true for professional use cases, the RAM difference is enormous and not usable for complex simulations. Same about GPU support that's used in various ML use cases. Or do you think you can train neural networks on M2 Ultra? If Apple was serious about actual professionals (not just artists), they would work with NVIDIA and others on this machine.

We're going to have to agree to disagree on this front, and it appears Apple does as well.

 

[NT1440]

The GPU is one confusing part of this calculus, but I think the bigger loss is the inability to upgrade the RAM. This is a core tenant of truly professional computers as computing requirements scale up. I also think it runs completely counter to the “environmental“ goals of Apple, as it makes it much more likely for a computer to end up in a dead-end e-waste situation because it can‘t be reconfigured to a new use-case.

I think you guys need to understand that certain workflows benefit massively from Apple’s GPU being able to tap into 100+ GB of RAM.

The studio/Pro can do things that are literally impossible on other devices because if this.

Conversely, workflows that rely on MASSIVE data 200+ GB) sitting in RAM are not suited for these devices.

 

[Alan Wynn]

We need exactly this, and have ordered fair quantity of the new Mac Pros to replace our current Mac Studios with external PCIe chassis.

We ordered 4 at launch, and expect to get more this week.

We use multiple SDI/SMPTE2110 video capture/output cards, 25Gb/100Gb network cards and NVMe HBA cards, all of which are crippled to varying degrees by the 4-lane Thunderbolt bus, and have been waiting for exactly what the new Mac Pro provides.

For us it is Blackmagic Hyperdecks and 100Gb/s Ethernet, as well as some multi-port 25Gb/s Ethernet.

We're hardly a fringe case either; from speaking with friends in the industry, everyone using MacOS is buying them up. The Mac Studio is fantastic, but PCIe support is crippled; the Mac Pro fixes this.

Exactly.

You're not the target market, and fail to imagine a use case. Real-world customers are buying the product while you complain on the internet.

Yup.

 

[Longplays]

We ordered 4 at launch, and expect to get more this week.

For us it is Blackmagic Hyperdecks and 100Gb/s Ethernet, as well as some multi-port 25Gb/s Ethernet.

Exactly.

Yup.

But what about it not being as strong and brave as a i9 or RTX 4090?

 

[CarAnalogy]

We need exactly this, and have ordered fair quantity of the new Mac Pros to replace our current Mac Studios with external PCIe chassis.

We use multiple SDI/SMPTE2110 video capture/output cards, 25Gb/100Gb network cards and NVMe HBA cards, all of which are crippled to varying degrees by the 4-lane Thunderbolt bus, and have been waiting for exactly what the new Mac Pro provides. We're hardly a fringe case either; from speaking with friends in the industry, everyone using MacOS is buying them up. The Mac Studio is fantastic, but PCIe support is crippled; the Mac Pro fixes this.

You're not the target market, and fail to imagine a use case. Real-world customers are buying the product while you complain on the internet.

Great, thanks for the feedback. I’m not the target market and it doesn’t make sense to me on paper but I’m glad to hear from someone that understands. I’m genuinely surprised to hear that you find the PCIe slots to be worth the premium.

 

[AlphaCentauri]

That’s what I’ve been saying previously. New Mac Pro is aimed at pro customers with very specific needs for PCI expandable Mac (not 3D rendering Mac or science/I need 1.5TB of ECC memory Mac).

Apple decided to abandon traditional 3D workstation market otherwise. It’s their measured decision and users who are not happy should vote with their wallets by swapping to Win/Linux workstations, instead of crying out their tears on internet forum 🤣

 

[KilgoreTully]

Apple is a trillion dollar company. Their schtick might be making slick technology, but their driving purpose is making money, and they are exceptionally, superlatively good at making money. Apple is not as good at making computers as they are at making money, which is not to say their computers are bad, per se; they're just that good at making money.

The whole "ecosystem" thing certainly has a lot of easily-touted and more-or-less legit consumer benefits, but the value to Apple is in cranking up the odds that customers buy more Apple products. Is the money-gobbling corporate juggernaut going to sell the best possible computer, that users can optimize by buying other companies' stuff, or develop an intriguing, niche alternative that locks in deep-pocketed users?

It's not a coincidence or scheduling fluke that Bob Iger came on stage. That was a show for all the shareholders and creative execs out there. There is now the impression, regardless of how accurate it is, that the most valuable entertainment brand in history does its tech stuff with Apple products. You want to make a billion dollars? Better listen to Cook and Iger! Other execs who are more focused on market perception and making money than understanding tech are going to buy Mac Pros by the pallet. Will all their devs/techs/artists be happy? Maybe (they're still beastly systems). Will it be a smart investment? Very maybe (abysmal upgradeability). Will Apple have accomplished their goal? Yep (Capitalism, baby!).

Sure, there are companies (and individuals) out there that need to customize some absolutely monster rigs, and CTOs who do understand their own highly specific computing needs, but Apple simply isn't in the business of making components for custom supercomputers that can model the atoms in a nuclear explosion or global weather patterns or whatever. If your AI research requires Nvidia GPU architecture and TBs of RAM, Apple isn't shunning you; they're just tacitly acknowledging they aren't trying to seriously move on the current ML market with their system architecture.

Bottom line, you don't have to like what Apple is doing with the Mac Pro, but it's just silly to say they are messing up. They know what they are doing. Their strategy is very clear. They want to make every part of the products they sell and appeal to the premium/luxury market. This has worked extraordinarily well for them, by the metrics that matter to them, and this is not a huge deviation from that winning strategy.

 

[nylonsteel]

cmon aapl seriously - for a machine with expansion slots and is top of line product - let there be graphics card support - ie nvidia
reminds me of my mac mini with old intel udh 630 graphics - non upgradable
though i shouldnt complain too much - i like my core i7 mac mini in space grey just fine
feel sorry for the mac pro peeps

 

[flottenheimer]

Prediction: Apple will kill the Pro lineup sooner rather than later.

It will go something like this:
"Pro users have simply been blown away by the incredible performance of the Mac Studio, with almost no one opting for the Mac Pro. With todays introduction of an even more powerful Mac Studio in a stunning Space grey color, we..."

 

[huge_apple_fangirl]

Does Apple care about Mac Pro users who want 2019 features returned to the 2023? It will be a net LOSS to Apple.

AMD/Intel/Nvidia are better situated to service them. It is a net gain for Apple to not sell to the vocal minority.

The Mac Pro has always been a niche device that, if you consider the opportunity cost of development, probably never broke even. The point was to make Macs capable at the highest end, so people who needed it stayed in the ecosystem and Mac stayed as an option. This strengthens the rest of the Mac lineup by having more options and software support. It’s not about the Mac Pro individually, which never made sense on its own.

 

[Longplays]

The Mac Pro has always been a niche device that, if you consider the opportunity cost of development, probably never broke even. The point was to make Macs capable at the highest end, so people who needed it stayed in the ecosystem and Mac stayed as an option. This strengthens the rest of the Mac lineup by having more options and software support. It’s not about the Mac Pro individually, which never made sense on its own.

Hence the Mac Studio that cuts down the price to $2k & $4k for the base models instead of $7k or more for I/O and PSU that less than 1% of Mac use cases have need of.

 

[deconstruct60]

Remember, Apple chose PCIe 3.0 for the Mac Pro 2019 even though PCIe 4.0 was released in 2017. PCIe 3.0 was released in 2010.

Really not truthful. Apple did not design the CPU or PCH packages for the Mac Pro 2019. Intel did. So Intel choose PCI 3.0; not Apple. Apple just worked with the design decisions that Intel made.

Intel did not release a W-2xxx or W-3xxx class CPU package with PCI-e v4 until 2021. There is NO WAY Apple could 'choose' something that was going to ship until 2021 in 2018-2019 when the Mac Pro 2019 had to be finalized.

What Apple shipped in 2019 is exactly what was a current Intel product in 2019.

The only relatively rapid adopter of PCI-e v4 was IBM ( Power Series and mainframes). There is a vast mulitple year gap where EVERYBODY in the PC industry did nothing. Intel , AMD , etc. It was not an 'Apple' thing at all.


Late 2019 AMD shipped some Eypc server processors with PCI-e v4.

AMD Rome Second Generation EPYC Review: 2x 64-core Benchmarked

www.anandtech.com

And the Threadripper 3000 series that was based off that core product set.

Pragmatically, those were not a real option for Apple.

1. AMD couldn't supply solutions for the rest of the 2019/2020 Apple product line. Forking Mac Pro off from the rest of the Mac line up was not that path that Apple was on. Especially, a year away from the transition off of x86_64 anyway! ( 2019 and the intel macs that slid into early 2020 were 'last gasp' x86_64 models. Not some new generation for the long term future. )


2. AMD over the 2013-2018 time span was relatively hostile to Thunderbolt. Thunderbolt was a key attribute of the Macs by 2017-18 when the design for Mac Pro would have been mostly laid down.

3. Back in 2016-18 AMD was just trying to keep the lights on. Yes, Apple was buying GPUs from them , but those were not being crisply delivered on time. ( Remember the Mac Pro shipped with a relatively 'old' 580X as the basic card because there wasn't a newer viable alternative at that price point in AMD's line up. And the W5700 was a 'paper launch' for the MP in 2019... didn't show up until 2020. ) Doubling down on 'maybe AMD will pull it off' probably was not a notion that someone was willing to beat the farm on.


4. Threadripper had 'max' core count feature 'check box' , but it was not better at single threaded. It was very , very ,very much so a multiple user , multiple conncurent 'different programs' optimized chip than a workstation one. It fix some corner cases, single user wasn't the focus.

The line up was also short. 24 , 32 , 64 cores . And the 64 solution was a 2020 product also.

https://www.anandtech.com/show/15318/amds-64core-threadripper-3990x-3990-sd

[ AMD's high priority was shipping Eypc models , not TR ones]

Not that 64 would be useful since macOS has a 64 thread limit. So would likely not to turn off SMT for that model ( yet another mismatch in the systems design).



If Intel had delivered the W-3300 in late 2019 with decent thermals [ like probably most 2016-17 roadmaps said it was going to do.] then Apple probably would have taken it. Intel didn't. So the PCI-e v4 , 2019 option only happened in an alternative universe.

They highlighted the 2023 Mac Pro having 2x the PCIe bandwidth over the 2019.

Apple highlighted the theoretical bandwidth to the slots is higher. Apple didn't highlight at all cleanly what the actual backhaul bandwidth is to those slots.

None of the PCIe peripherals [other than SSD/NVme based 5.0 PCIe cards fully leverage 5.0] cited would need more than PCIe 3.0 to use since either they are Audio or Data Array only based cards.

But could you really get the PCI-e v3 on the six slots in the MP 2019 ( slot 2 , 4 , 5, 6-8) . The backhaul to six of the PCI-e v3 slots in the MP 2019 is just two x16 PCI-e v3 lane bundles. There is are two x16 , x8 , x4 and some MPX double x4s all attached to the back hual. If wanted to fill those two x16 slots and run them full blast the other stuff in slots 6-8 and any MPX connector used wouldn't have any bandwidth left.

The biggest miss [deliberate omission] was not supporting CXL 2.0 and folding that into their designs for the future.

CXL 2.0 largely is a band-aid to try to get what Apple has with their "Unified Memory" approach. It probably would be useful to have eventually for M-series, but back in 2017-2018 CXL wasn't the completely unified front it is now. Intel came up with the foundation of CXL and lots of vendors (including AMD) fought them on it before they submitted and got on board. AMD even delayed Zen 4 Eypc a bit to 'change horses' and get on board. But a small part of that though was Apple doing what they are doing. ( promoting the 'extra copying is bad' and leads to RAM purchase bloat movement. )

 

[HurtinMinorKey]

"and 3D simulations"
Maybe for straight physics sims, but Apple's 3d graphic rendering performance without a real GPU will be absolute ******* in this setup. It's basically going to be on par with NVDA's desktop GPUs from 6+ years ago.