Mac Pro Enthusiasts Raise Concerns Over Upgrade Limitations of Apple Silicon

[Unregistered 4U]

Just the wrong terms to use, maybe potential or returning Mac Pro buyers have concerns. Yes I think the title of the thread is wrong.

Potential or returning buyers don’t have concerns. They currently have a system that’s fine for them and they’ll use it until it’s no longer suitable. Prior to it dying, they’ll look at the market and buy whatever suits their needs. It’s the enthusiasts, that likely won’t even BUY whatever Apple releases, that don’t want to be chided by their non-Apple friends about the “Mac that doesn’t have expandable RAM!”

The potential buyers stopped hanging around those “friends” a looooong time ago.

 

[Unregistered 4U]

I think the Mac can be in trouble in to long run if the real pros does not get competitive performance on the mac pro.

The Mac could be in trouble if the mobile systems they sell a majority of stop selling well. Mac Pro’s sell less than a million a year, perhaps even FAR less. If their sales crater, Apple will still end the year selling somewhere between 20-30 million Macs.

 

[Realityck]

Potential or returning buyers don’t have concerns. They currently have a system that’s fine for them and they’ll use it until it’s no longer suitable. Prior to it dying, they’ll look at the market and buy whatever suits their needs. It’s the enthusiasts, that likely won’t even BUY whatever Apple releases, that don’t want to be chided by their non-Apple friends about the “Mac that doesn’t have expandable RAM!”

The potential buyers stopped hanging around those “friends” a looooong time ago.

Enterprise will retain preconfigured workstations for usually 4 years, their primary concern is being able to easily run existing software licensing on what they configure/purchase. In other words the expense of software licensing exceeds the cost of the hardware. All the workstations are sold to a third party to sell them rather than retain them beyond the point that obsolesce or hardware failures occur. This is how it works no matter if you're talking about companies with large Mac or PC usage.

 

[Unregistered 4U]

Enterprise will retain preconfigured workstations for usually 4 years, their primary concern is being able to easily run existing software licensing on what they configure/purchase. In other words the expense of software licensing exceeds the cost of the hardware. All the workstations are sold to a third party to sell them rather than retain them beyond the point that obsolesce or hardware failures occur. This is how it works no matter if you're talking about companies with large Mac or PC usage.

And, those are also the potential buyers that are not concerned one bit about “memory being locked down”. It’s effectively locked down (by the contract between the companies) before they get it anyway. Fiddle around inside it and the user might void the protection service they’ve paid for when something happens in the future.

 

[Dydegu]

An M2 Mac Mini beats the Mac Pro in performance. So the days of most professionals shelling out 50K for a performance Mac are over.

Hot take, but with the advent of the Mac Studio, the Mac Pro should just be discontinued.

 

[Realityck]

And, those are also the potential buyers that are not concerned one bit about “memory being locked down”.

This is just continuing the debate about the necessity of socketed memory vs soldered unified memory, specifically asking for Apple to continue to imitate PC workstations rather than newer computer technology alternatives.
I await what Apple has to offer, hopping its more flexible and a lot less art-gallery in appearance.

 

[Detnator]

But will it blend?

 

[Nugget]

But will it blend?

Sure.

 

[cupcakes2000]

They could add the ability to just add extra mX cards in. So the base config could have, for example, 6 extra free slots to purchase extra complete max modules which are stackable. 6 stacked maxed out m3 ultra modules would certainly cost a bit, but what a machine!

 

[deconstruct60]

The Mac could be in trouble if the mobile systems they sell a majority of stop selling well. Mac Pro’s sell less than a million a year, perhaps even FAR less.

Mac Pro's are no where even close to 1M. 0.1M maybe, but even back in their hey day they were never in the 1M/yr range. Increase the entry price by 100% and they there isn't going to be a million unit stampede. Apple said Mac Pro was in the 'single digits' percentage of Mac market. Some people try to spin that into 7-8% but rounded up probably closer to 1% .

5-6 years at 100K per year is 500-600K folks. Still not a 1M, but a sizable group if relatively slowing buying new ones group.


If they combine the Mac Studio + MP , then probably could get into the 1M/yr ballpark range. The more of the Max range loop in the higher the likelihood.

If their sales crater, Apple will still end the year selling somewhere between 20-30 million Macs.

Apple sales are only in the 20's M (measuring the pandemic mania buying levels as some 'norm' is dubious.)

" Apple 6.3 (million) 22 5.8M (21 ) ... "

Apple's Mac Shipments Grew in Q2 2022 Amid Continued Worldwide PC Shipment Decline

Apple's worldwide Mac shipments were up in the second quarter of 2022, according to new PC shipping estimates shared today by Gartner. Apple...

www.macrumors.com

4 * 6.3 25.2M ballpark.

If actually 'cratered' from there , then likely falling below 20M. [ Not likely as Apple can just toss some profit margin out and keep some volume offsets. ( 3-4 month discounts on M1 MBP 14"/16" while sat on them before rolling out upgrades. ]

 

[deconstruct60]

They could add the ability to just add extra mX cards in.

There are not interfaces on the plain Mx SoC to connect to anything else in a reasonable high bandwidth/low latency fashion. Even the M2/M1 Max have only four 1x PCI-e v4 lanes of general usage bandwidth. Put a x1 PCI-e card into a slot and get what in terms of a cluster node interconnect support?

To get to the much higher than 'normal' memory bandwidth levels of the M1 and M2 series Apple threw a much higher than 'normal' amount of the die edge space at memory controllers than most general PC SoC vendors do. The dramatically shrank the aggregate external PCI-e bandwidth versus a Gen 10 Intel SoC/Chipset used a Mac and repurposed that to much higher memory bandwidth. With that trade off Apple iGPUs are crushing Intel's. But there was a 'price' for that. The normal M-series SoC is only primarily in talking to external memory chips and not to anything else.

You can try to handwave at "UtlraFusion" , but it only useful for extremely , hyper short distances. Not a card several inches away in a physical plug in socket.


At best, you need some new SoC that Apple added some longer distance I/O to with larger bandwidth and tolerable latencies ( e.g. x16 PCI-e v4/5 )

So the base config could have, for example, 6 extra free slots to purchase extra complete max modules which are stackable. 6 stacked maxed out m3 ultra modules would certainly cost a bit, but what a machine!

Stackable? If the Max module is 80W how are you going to 'stack' another 80W module on top of that? The M1 Ultra package places the Max dies side-by-side precisely because you cannot stack that in a reasonable way thermally.

If somehow referring to the vertical arrangement of the slots on a vertiable 6 slot motherboard , those are not really 'stacked' at all. If you take the whole board/system and rotating 90 degrees and rest the system on it's side those slots are not 'stack' in any sense of the word.

The M1 Max dies in a M1 Ultra package are 'stacked' on an interposer. If adjust the logicboard that Ultra package is on from horizontal to vertical then they are still just as 'stacked'. ( e.g., flip a M1 Ultra Mac Studio on its side. )

Six Ultra logic boards inside of a single container box. ( so all six computers share a single power supply ) might be interesting for some Mac cloud services vendors, but fairly likely that most single user Mac Pro users will be cheering for that.

 

[DeepIn2U]

Maybe at one point in the distant past. The majority of light mobile computers that folks are actually purchasing has no allowances for ‘growing to needs’ except for if the OS does more or a new app that does a thing someone needs comes along.

That’s exactly what I meant.

Only recently has iPadOS and iPads got apps that really needed the RAM boost that only the M series chips could provide, along with processing power. Will definitely see more in a few short years.

 

[DeepIn2U]

They’re both x86, both trying to work around an outdated infrastructure that requires them to pump more power in in a way that doesn’t even provide a proportionate performance increase. PLUS, while they tout their performance as compared to Apple Silicon (who ever thought that would be a thing that AMD would even tout!!), they know Apple’s not their competition. Meeting the performance/power efficiency of Apple doesn’t matter to them because it’s not like Windows users have an option of using Apple Silicon.

Don’t forget that both, moreso by Intel, NEED to have their x86-64 CPU’s support legacy code for business applications.

This is why more corporations still predominantly order computers that have Intel cpu’s over AMD.

In a few short years, 3, we’ll finally see developers creating ARM based applications and hopefully have better porting for macOS ARM CPU’s than ever was the case with x86-64 on both platforms.

Games, the top tier that require NVidia and AMD GPU’s May finally be available, maybe even performance tweaked, for MacOS.

I know I’m dreaming here but hey it’s more possible than ever before

 

[falainber]

They’re both x86, both trying to work around an outdated infrastructure that requires them to pump more power in in a way that doesn’t even provide a proportionate performance increase. PLUS, while they tout their performance as compared to Apple Silicon (who ever thought that would be a thing that AMD would even tout!!), they know Apple’s not their competition. Meeting the performance/power efficiency of Apple doesn’t matter to them because it’s not like Windows users have an option of using Apple Silicon.

You are aware that ARM is almost as old as x86, right?

 

[falainber]

Don’t forget that both, moreso by Intel, NEED to have their x86-64 CPU’s support legacy code for business applications.

This is why more corporations still predominantly order computers that have Intel cpu’s over AMD.

In a few short years, 3, we’ll finally see developers creating ARM based applications and hopefully have better porting for macOS ARM CPU’s than ever was the case with x86-64 on both platforms.

Games, the top tier that require NVidia and AMD GPU’s May finally be available, maybe even performance tweaked, for MacOS.

I know I’m dreaming here but hey it’s more possible than ever before

In 3 years, x86 ecosystem will still have orders of magnitude more business applications than ARM. Support for legacy code is paramount for business.

 

[falainber]

An M2 Mac Mini beats the Mac Pro in performance. So the days of most professionals shelling out 50K for a performance Mac are over.

Hot take, but with the advent of the Mac Studio, the Mac Pro should just be discontinued.

M2 Mac Mini beats Mac Pro only in single core performance. That would be important if the computers still had single core processors. They don’t. With all the other performance factors considered, M2 Mac Mini is still a computer toy compared to real workstations.

 

[Mago]

There are no Technical Challenge for Apple to deliver a Quad M2 Max Mac Pro, IMHO Gurman moved to clickbait business model.

RAM:, Until now all Mac with M1/M2 chips use Soldered Ram, indeed dont need RCDs to interface with DDR5 Dimms, apple just add this component and m2 Max not restricted to "in die" ram (actually not in Die, as are pcb soldered) as the RCD enable it CXL-PHY jedec-std DDR5 connections, but there is an catch, the Mac Pro likely to be very picky on DDR5 modules, and a minimum 2 channel by each M2 Max on board shall be populated (from 4) thus the Mac Pro likely will require 4 DIMM for M2 Ultra, and 8 DIMM for M2 Extreme all populated with same kind, size, cas, etc dimm modules, and likely upto 16 DIMM could connect, 16x48GB =768GB using consumer grade DIMM, with commercially available (and crazy expensive) 256GB server modules the Mac Pro could be configured with upto 4TB of ram, buy likely you'll need to sell your Lambo to pay for it likely north 80K$.

PCIe expansion: the Fabric Interconnect Chip (aka Ultra Fusion) provided for a PHY/PCIe interface which Apple May configure without restrictions, but I believe it will provide only 2 PCIe5 x16 slots and 2 or 3 PCIe4 X16. or 2 PCIe5x8 + 1 PCIe 4x16, PCIe5 slots enabled to install Compute Accelerators including GPUs.

Discreet GPU: M1/M2 mac cant connect external GPU using thunderbolt even booting Linux due Apple (I think as safety meausre) restricted Thunderbolt PCIe what is known as DMA or direct memmory access, what enables an peripheral to write data back to system RAM without occuping the CPU, it also saves integration with PHY/Fabric, but nothing Apple Engineers not aware how to enable on STD PCIe devices (if I own an M2 Max MBP i'll try now to test if it recognizes TB3-linked eGPU).
About the Software, Mac OS and Metal never removed support for multiple vendor GPUs as long those GPUs load the respective platform compatible Drivers, so from the respective Driver Update the Mac Pro should be capable to use AMD CDNA2/RDNA2 even CDNA3-RDNA3 GPUs as the monster Instinct MI100 or the productive (and more powerful than M2's GPUs) RX 7900 XTX, maybe even later intel could join to the party with their ARC gpus, since Intel isn't banned at apple neither they ban apple as nVidia.

M2 MaX AND Waffer Availability: Gurman argues that Apple is discarding the M2 Extreme as they want to preserve M2 Max waffer yield for MacBooks, minis, and Studios... that's not even coherent from someone serious, is know Apple cut M2 Macbook orders from suppliers due weak demand, how an 8000-12000$ (probable base price for M2 Extreme Mac Pro) could eat as many SOC to compromise Macbook Production, IMHO apple likely Apple will have M2 Max to spare.

My predictions:

Early March Apple should Release the M2/M2 Pro iMac 24 along show the VR/AR glasses, the at WWDC after M2 Max/Ultra Mac Strudio Introduction should introduce the M2 Ultra/Extremme Mac Pro, sharing the iconic CheeseGrater chassis (even with 400$ wheels), with upto 4TB ram, 16TB NVME, single/dual AMD GPU optional, starting somwhrere near 6000$ for base M2 Ultra/64gbDDR5 ECC DIMMS /1TB and close 12000$ for M2 Extremme Models starting with 128GB DDR5 ECC DIMMS, none including dGPU, and options based on newest radeon GPUs priced as current ones, old MPX likely modules not compatible with the new Mac Pro as it likely to ditch the x16 PCIe extension it requires, but other MPX modules not occuping the PCIe extension likley compatible from Day 0 as HDD cages, even dont discard and Afterburner successor, based on ASIC instead FPGAs indeed ridiculous powerful bu at least 20x the current card, or using half power on an 10x more powerful accelerator.

I've been tiped (this is just speculation, take with tons of sea salt), along the Afterburner, apple to introduce an monster TPU/NPU accelerator card, which will bring the Mac Pro the crown as the most powerful AI training workstation.

Gurman, if he don't have actual news shouldn't fabricate it, much better for everyone.

 

[SouthPalito]

You clearly haven't been to any music recording studios.

I'm sure people enjoy using logic. Still, that's not enough business to support the development and sales of a very expensive workstation that does not have the power of Linux/windows solutions that offer the fastest CPU GPU TPU technology and sell in huge volumes.

 

[mode11]

RTX4090 is a SoC by itself. It contains a CPU a GPU and a ram system.
All Nvidia "GPU" since RTX2000 are SoCs with arm core on it for communicating with driver.

What ARM core? A single A77 or something? Not really comparable to an M2 CPU cluster, plus all the other logic for TB etc.

 

[Alan Wynn]

In 3 years, x86 ecosystem will still have orders of magnitude more business applications than ARM. Support for legacy code is paramount for business.

Nah. Support for MS Office and the web are paramount for business. Very few companies run custom applications that are native code these days.

 

[falainber]

Nah. Support for MS Office and the web are paramount for business. Very few companies run custom applications that are native code these days.

For this (web) type of applications, nobody is going to buy Macs. $300 PCs will do just fine.

 

[zakarhino]

Sure, I can definitely understand why people might be disappointed with the lack of upgradeability in the new Mac Pro, especially since the current Intel-based model is so popular with creative professionals due to its modular design. However, I do think there are some potential upsides to this move. One thing that comes to mind is the possibility of using the PCI slots for domain-specific cards. For example, you could add an AI processor for more advanced machine learning tasks, a ProRes processor for faster video encoding, or even a third-party GPU with RTX support for even more powerful rendering capabilities. And who knows, maybe the built-in GPU will be able to work in conjunction with these third-party GPUs for even more power. Additionally, the PCI slots could be used for Network cards and NVME SSD's which are also important for media production and other professional workflows. It's definitely an interesting development and I'll be looking forward to seeing how things pan out with the new Mac Pro.

 

[Blowback]

The new Mac Pro coming later this year is expected to feature the same spacious modular design as the 2019 model, but with fresh concerns over its lack of upgradeability surfacing, some users are beginning to wonder what the transition away from Intel architecture actually means for Apple's most powerful Mac.

The current Intel Mac Pro that Apple sells is popular with creative professionals because of its modular, highly upgradeable design. It accepts Xeon processors with up to 28 cores and 64 PCI Express lanes, and up to 1.5TB of high-performance memory. It also has eight PCIe expansion slots, while the APX Modules can be configured with six different GPUs. All of the components can be upgraded after purchase.

Given that Apple's high-end M2 Extreme chip has reportedly been cancelled, the new Mac Pro is expected to feature a new M2 Ultra chip, which is essentially two ‌M2 Max‌ chips connected together,‌ featuring up to a 24-core CPU, up to a 76-core GPU, and at least 192GB of unified memory.

But it is just this unified architecture that is raising doubts about the Mac Pro's modularity among prospective owners. Earlier this month, Bloomberg's Mark Gurman revealed that unlike the current Intel-based ‌Mac Pro‌, the upcoming model will not feature user-upgradeable RAM because the memory is tied directly to the M2 Ultra's motherboard.

Now, Gurman has seemingly doubled down on this limitation of Apple's unified architecture by suggesting that the next Mac Pro "may lack user upgradeable GPUs" in addition to non-upgradeable RAM. As he points out: "Right now Apple Silicon Macs don't support external GPUs and you have to use whatever configuration you buy on Apple's website."

https://twitter.com/x/status/1618535843203014657

https://twitter.com/x/status/1618591311950667777

Apart from space for media and networking cards, that leaves only storage as the main user-upgradeable component in the Mac Pro. As for the seemingly large empty space that would remain in the unchanging tower case, Gurman suggests it could accommodate a larger cooling system, which would differentiate it from the Mac Studio by affording significantly faster performance.

It's worth noting that on the subject of non-upgradeable GPUs, Gurman appears to be reaching this conclusion based on logical reasoning rather than specific insider information. But if it turns out that he is correct, it could dissuade would-be buyers from investing so much money in a machine that does not offer the same future-proofing expandability.

The ‌Mac Pro‌ currently starts at $5,999, with the potential to price it up to over $52,000 in custom configurations. The base model Apple silicon ‌Mac Pro‌ with the ‌M2‌ Ultra is almost certain to cost more than the high-end ‌Mac Studio‌ with the ‌M1 Ultra‌, which starts at $3,999.

https://twitter.com/x/status/1618579444024868867

Apple has already teased the launch of the first Apple silicon Mac Pro, so it is undoubtedly coming regardless of the rumors. Indeed, Apple is already testing one of the new ‌Mac Pro‌ models, and it is running macOS Ventura 13.3, an updated version of ‌macOS Ventura‌ coming out later this year. Timing wise, ‌macOS Ventura‌ 13.3 is expected to come out around March or April, so we could also see the ‌Mac Pro‌ launch in this timeframe.

Article Link: Mac Pro Enthusiasts Raise Concerns Over Upgrade Limitations of Apple Silicon

As of now: All speculation. No one knows ...or WILL KNOW..until Apple releases its product. End.

 

[sir1963nz]

Hasn't it been studies showing that peo0ple seldom upgrade RAM. So buy the amount that is a bit above your current maximum need.

I can understand the graphic card, if those change a lot. But my experience with the m-series is that they seem to compete very well with any other card. So maybe it is not needed to change your GPU every six months?

What I really hope and wait for is storage. I want local storage and a nice case-design, so I can use it for daily use now and then later retire it as a server just as I did for my mac pro 2010.

Until my current machine, I have bought the minimum I needed and the bought extra RAM/Storage as my needs grew.
I have done this since my first TRS-80 clone back in the 1980's.

I have 2 Mac Minis that I run as servers at home, one a 2009 the other a 2011.
I have about 50TB of storage attached to them, and I have upgraded the RAM in them and replaced the old HD with a much faster bigger SSD.

I am looking at replacing these with 2nd hand 2014 models, not new.

 

[T-R-S]

Having owned a Studio Ultra with 128GB of RAM and a 8TB SSD with the high end graphics it’s got more jam than I use. With 2 5K 100 MHZ displays it’s so powerful. Must people don’t even need the current video power as it is. How many people here need to run 6 - 8K displays?