Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
What's Happening With the Mac Pro?
- Thread starter MacRumors
- Start date
- Sort by reaction score
Currently, Apple hardware does not offer a choice. You cannot run native Windows or dozens of Linux distributions. It is unclear what the situation will be in a few years. With Hackintosh, you have a choice; you can do anything, but you don't have to.
I honestly could see initial M2 Ultra Mac Pro sales being "robust" for anyone using a 2019-era Mac Pro for non-GPU accelerated tasks since it was clear Intel support in macOS was not long for the world. But once that initial demand was met, sales of the model fell off a cliff since "everyone who needed/wanted one had one".
I expect Apple could make a PCIe card with thousands of GPU or NL cores on it, coupled with terabytes of RAM, but without third-party support, how useful would it be? Even if Apple offered CoreML support for it, the common refrain in these forums and elsewhere is "It's CUDA or Nothing!" so how many people would spend the astronomical amounts of money it would cost if it only supported Apple's AI?
If Apple ever updates the Apple Silicon Mac Pro, you might be able to do the same (provided you can secure a system board).
And more problems created since you increase cost and complexity. Back in 1986, my $25,000 HP Vectra RS/25C workstation had its 25MHz 80386 CPU, 80387 FPU, 82385 Cache Controller (effectively an external L1 cache for the CPU) and 16 30-pin DIMM slots on a daughter card that plugged into the system board backplane that had the ISA slots for peripherals. I presume this was designed to allow customers to start with a slower CPU (like the 16MHz model) and then upgrade to a faster model (like the 33MHz model) without having to toss the entire chassis.
But once the i486 was released, HP never offered such an upgrade and I instead had to replace my i386 with a Cyrix 486DLC. However, even if HP had offered an i486DX option, I would still have been stuck with 8-bit and 16-bit ISA for expansion, missing out on PCI and VESA expansion slots to support faster video, storage controller and network controller cards.
So even if Apple offered an M6 system card to replace the M5 card the machine originally came with, if the new M6 model came with faster PCI or TB generations that allowed for faster SSD speeds or newer peripherals like HDMI 2.2 or 100Gb Ethernet ports, you would lose out on all that. Now perhaps you personally would not care, but presuming that M6 card will probably cost over 50% of the price of the new M6 model, how many would just not buy a new machine? Or wait until the M7 or M8 arrives with even more performance and replace the entire machine then?
Last edited:
Apple bringing high end single threaded performance to even the lowest end devices is “dropping the ball”?
Does a 32-core CPU really need a second socket? The 64 cores in the Threadripper Pro 9985WXs are barely faster in practice.
Depends entirely on what you’re doing. There’s plenty of “embarrassingly parallel” workloads where that absolutely helps
...but that doesn't mean that there's a market for a Mac with those specs. Those "multiple high-end GPUs" (probably NVIDIA or AMD) aren't going to magically get faster because they're slotted into a Mac rather than a HP. In fact, they're gonna get slower because (even if Apple added PCIe GPU support) a Mx Ultra only has enough PCIe lanes to drive a single 16x GPU. Most must-have software at that level is available for Windows and/or Linux, the "user friendly" Mac UI is no big deal if you're immersed in hugely technical pro software, and the power efficiency of Apple Silicon is drowned out by those big, sweaty GPUs.
The reality is that Apple Silicon is great for the tablets, laptops and small-form-factor systems that account for the vast majority of Mac sales but if you want a multi-PCIe-GPU workstation, a M-series chip just isn't the tool for that job. Apple doesn't have anything special to bring to the "big box'o'slots" market - and they don't need to.
The only advantage of that would be if it used "raw" PCIe for RDMA interconnects to get an edge over Thunderbolt. An Mx Ultra still only has 16-24 lanes of PCIe to play with, but that's usually a generation ahead of Thunderbolt. Not sure how much that would improve performance, and they'd need to design a 5xM5 Ultra logic board with all the PCIe plumbing, and a case... to serve a tiny market.
What they have with the Studio is something that performs exceptionally well in relation to its size and power consumption - even if it can't compete with a rack stuffed with HP's finest - with a (relatively) large potential market, with a software-only (well, plus a bunch of Thunderbolt cables) route to scaling it to multiple machines for more specialist apps. 4 Studios in a stack is still smaller and more power-frugal than a massive tower - and there's plenty of scope for a third-party to come up with a better way of organising it than a generic mini-rack designed for Raspberry Pi clusters (as shown in the video) and regular TB5 cables.
It just won't though, because you can't fit half a a few hundred gigs on your low end Nvidia card's VRAM.
There aren't that many where 1) many cores helps, 2) I/O is not a bottleneck, and 3) moving to GPU cores instead isn't practical.
This was a goal of Steve Jobs. It was one of the reasons when he returned to launch the G4 Cube to see how his idea would fare. He never liked anyone to have the ability to change out the internals of his machines. Apple Silicon favors his "closed system" design that he envisioned the Macintosh would be in his eyes.
Also, I knew Arm and RISCV based machines would need a different type of thought process when it came to upgradeability versus X86. Even adding in a basic M.2 slot on a Raspberry Pi required a daughter board. I hope the Mac Pro could evolve over time rather than the Studio replacing it.
Also, I knew Arm and RISCV based machines would need a different type of thought process when it came to upgradeability versus X86. Even adding in a basic M.2 slot on a Raspberry Pi required a daughter board. I hope the Mac Pro could evolve over time rather than the Studio replacing it.
Except its the niche machine that designs all the other machines...
But I do think it's a bit pointless with RAM / GPU baked into CPU.
I do wonder if they will create a bespoke connection to an enclosure that is faster and more Direct than Thunderbolt 5
A stacked system baed on Mac Studio.
Drive units
PCIE Units
Input / Output
GPU / ML units.
Mix and match - chose what you need to make it more powerful.
It isn't. That was true in the 1990s.
With ARM Macs, most of the design takes place on the iPhone, and most of it is just scaled up. Macs add some features like Thunderbolt, and complexity like dual-boot, but not that much. Apple does not take a high-performance machine and scale it down.
It would be another afterburner card. And look how useful that turned out to be when Apple shackled it.
Apple has to adopt standards rather than dogmas if it wants to do a workstation. And that means something else other than a king sized Mac Mini with an Apple Tax added.
That’s how they open it up to more buyers who aren’t tied to the Apple ecosystem.
Nice idea, it what if they have a change of heart and scrap the lot 2-3 years later for the next big thing? All of those become expensive hard to find orphans.
Try finding W6800X Duos or the W6900X card now at a reasonable price. You won’t. They are rare and expensive.
Half the Apple staff you speak to have no idea what a Mac Pro even is, they correct you “you mean a MacBook Pro” and then you need to explain what your computer is. That’s happened numerous times for me at flagship Apple stores and on their support phone lines.
Given that, why should you or anyone else risk buying such a product in the future?
Last edited:
Look at the current Mac Pro design. It would be a replacement logic board, so you’d get full support for all those things. Contemporary silicon architecture and advanced packaging doesn’t allow for anything else. Nvidia’s logic boards are no different.
But you’re right about the cost equation, and that’s likely why they won’t do it — a new replacement M2 Ultra Mac Pro logic board today is whatever it is, depending on memory and core counts, basically you can’t get it without a trade in. Between the logic board and new I/O cards, the cost of an upgrade kit would be at least 80% of a new machine.
Last edited:
PCIE Gen 6 does 1024Gbps vs 120Gbps of TB5. I could see a system with 4-6 Ai accelerator cards to really push the local LLMs.Not only that, the cluster functionality of Thunderbolt 5 seals the deal. https://appleinsider.com/articles/2...-boost-from-new-rdma-support-on-thunderbolt-5
In AI, only Nvidia matters, and Apple lost touch with the leaders long ago. Anyway, serious AI applications use computing servers whose power exceeds the imagination of the average PC or Mac user. AI in personal computers is just a toy.
Perhaps not, but (and this may blow your mind) other computers have upgradable memory as well as massive PCIe bandwidth. Max Studio maces out at 128GB memory which is pretty small.
Mac Studio with the M4 Max maxes out at 128, but with the M3 Ultra it maxes out at 512 GB.
Not true, GPU itself is extremely inefficient cause it was never meant for AI and compute. That's why NPU came out and Google's TPU proven that with Gemini 3.0 which outperformed OpenAI with Nvidia.
That's nothing compared to Mac Pro 2019 with 1.5TB of RAM and 128GB VRAM. And yet, Mac Pro 2019 had less RAM and VRAM compared to other workstations.
Gemini 3.0 it is software not hardware. Nvidia supplies 90% hardware for AI. You could say it has a monopoly )
But Apple can lock itself in its ecosystem and pretend it's the best at everything.
The architecture was designed for embedded systems first and it shows in the lack of expandability and repairability.
Only the very hardcore Apple fans that drank a bit too much of Apple Marketing's kool-aid believe otherwise.