As someone who came from the SGI O2/Octane era when high-end workstations were compact, distinctive, and sexy, I’ve never really understood the allure of the Mac Pro, with the exception of the 2013 Mac Pro tube, which I owned (small footprint, quiet, and powerful).
For me, aesthetics and size are important. That workstation on your desk should justify its presence, not just exist as some hulking box.
When Apple released the Mac Studio, it made perfect sense from a form-factor point-of-view. The internal expansion slots in the M2 Mac Pro didn't make any sense. It was like a bag of potato chips - mostly air. And far too big and ugly to be part of my work area! I'm surprised that Apple didn't discontinue it sooner.
As someone who worked on the M2 Mac Pro and has a real soft spot for it, I get it. It’s horrendously expensive and doesn’t offer much benefit over a Mac Studio and a thunderbolt pci chassis. My personal dream is that vms would support pci pass through and so you can just spin up a Linux vm and let it drive the gpus. But at that point, why are you buying a Mac?
SR-IOV and VFIO passthrough are different things. SR-IOV partitions a PCIe device across multiple VMs simultaneously (common for NICs and NVMe). VFIO passthrough gives one VM exclusive ownership of a physical device. For GPU compute you almost always want full passthrough, not SR-IOV partitioning.
The harder problem on Apple Silicon is that the M2 Ultra's GPU is integrated into the SoC -- it's not a discrete PCIe device you can isolate with an IOMMU group. Apple's Virtualization framework doesn't expose VFIO-equivalent hooks, so even if you add a discrete AMD Radeon to the Mac Pro's PCIe slots, there's no supported path to pass it through to a Linux guest right now.
On Intel Macs this actually worked via VFIO with the right IOMMU config. Apple Silicon VMs can do metal translation layers but that's not the same as bare-metal GPU access. It's a real limitation and I doubt Apple will prioritize solving it since it would undercut the "just use macOS" pitch.
They're trying to make it very clear they're not speaking on behalf of Apple Inc, despite having worked (or working) there.
Big companies like to give employees some minimal "media training", which mostly amounts to "do not speak for the company, do not say anything that might even slightly sound like you're speaking for the company".
It's dumb from a practical perspective. But I keep hoping they'll vertically compress their trashcan design so it looks like their Cupertino headquarters.
i wish i'd never traded in my 2016 mac pro (aluminum polished tube) as it was beefy, it was silent, clever thermo design (like the powerpc cube 20 years earlier or so), and i'd upgraded the living crap out of it for cheap.
I'm surprised they even tried selling an Apple Silicon Mac Pro - I expected that product to die the moment they announced the transition. Everything that makes Apple Silicon great also makes it garbage for high-performance workstations.
The allure of the Mac Pro is that you could dodge the Apple Tax by loading it up with RAM and compute accelerators Apple couldn't mark up. Well, Apple Silicon works against all of that. The hardware fabric and PCIe controller specifically prohibit mapping PCIe device memory as memory[0], which means no GPU driver ever will work with it. Not even in Asahi Linux. And the RAM is soldered in for performance. An Ultra class chip has like 16 memory channels, which even in a 1-DIMM per channel routing would have trace lengths long enough to bottleneck operating frequency.
The only thing the socketed RAM Mac Pros could legitimately do that wasn't a way to circumvent Apple's pricing structure was take terabytes of memory - something that requires special memory types that Apple's memory controller IP likely does not support. Intel put in the engineering for it in Xeon and Apple got it for free before jumping ship.
Even then, all of this has gone completely backwards. Commodity DRAM is insanely expensive now and Apple's royalty-bearing RAM prices are actually reasonable in comparison. So there's no benefit to modularity anymore. Actually, it's a detriment, because price-discovery-enforcing scalpers can rip RAM out of perfectly working computers and resell the RAM. It's way harder to scalp RAM that's soldered on the board.
Apple really dropped the ball here. They had every ability to make something competitive with Nvidia for AI training as well as inference, by selling high end multi GPU Mac Pro workstations as well as servers, but for some reason chose not to. They had the infrastructure and custom SoCs and everything. What a waste.
It really could have been a bigger market for them than even the iPhone.
Just about everybody who isn't Nvidia dropped the ball, bigtime.
Intel should have shipped their GPUs with much more VRAM from day one. If they had done this, they'd have carved out a massive niche and much more market share, and it would have been trivially simple to do.
AMD should have improved their tools and software, etc.
Apple should have done as you say.
Google had nigh on a decade to boost TPU production, and they're still somehow behind the curve.
Such a lack of vision. And thus Nvidia is, now quite durably, the most valuable company in the world. Imagine telling that to a time traveler from 2018.
I think for AMD, they were focused on competing against Intel. Remember AMD was almost bankrupt about 15 years ago because of competing against Intel. But the very first GPU use for AI was actually with an ATI/AMD GPU, not an Nvidia one. Everyone thinks Nvidia kicked off the GPU AI craze when Ilya Sutskever cleaned up on AlexNet with an Nvidia GPU back in 2012, or when Andrew Ng and team at Stanford published their "Large Scale Deep Unsupervised Learning using Graphics Processors" in 2009, but in 2004, a couple of Korean researchers were the first to implement neural networks on a GPU, using ATI Radeons: https://www.sciencedirect.com/science/article/abs/pii/S00313...
And as of now I do believe AMD is in the second strongest position in the datacenter space after Nvidia, ahead of even Google.
> And thus Nvidia is, now quite durably, the most valuable company in the world.
Nvidia is the most valuable company in the world right up until the AI bubble pops. Which, while it's hard to nail down when, is going to happen. I wouldn't call their position durable at all.
Don’t mistake stock market performance for revenue. NVIDIA makes ~200B annually, same as what Apple makes from iPhones. It’s a big market but GPUs aren’t just AI.
I'm purely talking in terms of revenue. There's a huge demand for AI systems from personal workstations to datacenter servers, and Apple was one of the few companies in the world in a position to build complete systems for it.
But for some reason Apple thought the sound recording engineer or the video editor market was more important... like, WTF dude? Have some vision at least!
> something competitive with Nvidia for AI training
Apple is counting on something else: model shrink. Every one is now looking at "how do we make these smaller".
At some point a beefy Mac Studio and the "right sized" model is going to be what people want. Apple dumped a 4 pack of them in the hands of a lot of tech influencers a few months back and they were fairly interesting (expensive tho).
> Apple is counting on something else: model shrink
The most powerful AI interactions I've had involved giving a model a task and then fucking off. At that point, I don't actually care if it takes 5 minutes or an hour. I've cued up a list of background tasks it can work on, and that I can circle back to when I have time. In that context, smaller isn't even the virtue at hand–user patience is. Having a machine that works on my bullshit questions and modelling projects at one tenth the speed of a datacentre could still work out to being a good deal even before considering the privacy and lock-in problems.
> At some point a beefy Mac Studio and the "right sized" model is going to be what people want.
It's pretty clear that this isn't going to happen any time soon, if ever. You can't shrink the models without destroying their coherence, and this is a consistently robust observation across the board.
I don’t think it’s about literally shrinking the models via quantization, but rather training smaller/more efficient models from scratch
Smaller models have gotten much more powerful the last 2 years. Qwen 3.5 is one example of this. The cost/compute requirements of running the same level intelligence is going down
Not all enterprises are the same, I imagine many companies have different departments working with local optimums, so someone who could benefit from it to get more productivity might not have access to it because the department that is doing hardware acquisition is being measured in isolation.
Give every iPhone family a in house Siri that will deal with canceling services and pursuing refunds.
Your customer screw up results in your site getting an agent drive DDOS on its CS department till you give in.
Siri: "Hey User, here's your daily update, I see you haven't been to the gym, would you like me to harass their customer service department till they let you out of their onerous contract?"
The Ultra variants of the M series chips had previously consisted of two of the Max chips bonded together.
The M5 generation Pro and Max chips have moved to a chiplet based architecture, with all the CPU cores on one chiplet, and all the GPU cores on another.
While the trash can generation was somewhat present and around, I don't think I ever saw a cheese grater in the flesh. Did it have any users? Were there any actual useful expansion cards? Did anybody continue buying this at all, after it didn't get the M3 Ultra bump, that the Mac Studio got last year?
It had many hardware upgrades over the years - upgraded CPUs, 128GB RAM, 4TB NVME storage, a modern AMD GPU, USB3/c, thunderbolt, etc
The only reason it got replaced is because it became too much of a PITA to keep modern OSX running on it (via OCLP)
Replaced with an M4 Max Mac Studio, which is a nice and faster machine but with no ability to upgrade anything and much worse hardware resale value on M-series I'll have to replace it in 2-3 years
I'm a former 4,1 user, myself — replaced with an M2Pro mini Jan 2023 (finally retired fully 2025).
Absolutely recommend you purchase the 4-bay Terramaster external enclosure — gives you four SATA slots that are hot-swappable (unlike MacPro's). 10gbps via USB-C.
The cheese grater mac pros were very popular, in that people got them and continued to use them.
The most notable feature was that there were mac-specific graphics cards, and you could also run PC graphics cards (without a nice boot screen). They had a 1.4kw power supply I believe, and there was extra pcie power for higher-end graphics cards. You could upgrade the memory, add up to 6 or more sata hard disks (2 in dvd slot). You could run windows, dual booting if you wanted and apple supported the drivers.
The 2013 was kind of a joke. small and quiet, but expansion was minimal.
2019 looked beefy, but the expansion was more like a cash register for apple, not really democratic. There were 3rd party sata hard disk solutions,
the 2023 model was basically a joke. I think maybe the pcie slots were ok for nvme cards, not a lot else (unless apple made it).
nowadays an apple computer is more like an iphone - apple would prefer if everything was welded shut.
Not surprising, as the market has broadly moved on from add-in cards in favor of smaller form factors and external devices, absent some notable holdouts in specific verticals.
Gonna miss it, though. If they had reduced the add-in card slots to something more reasonable, lowered the entry price, and given us multi-socket options for the CPU (2x M# Ultras? 4x?), it could have been an interesting HPC or server box - though they’ve long since moved away from that in software land, so that was always but a fantasy.
At least the Mac Studio and Minis are cute little boxes.
This makes sense, for that kind of money you could always build a beastly workstation in a real ATX case with standard components. Install Linux and the Mac looks like an expensive toy in comparison.
Still rocking a 2019 (Intel) Mac Pro here, all slots filled with various Pro Tools and UAD DSP cards, SSD, GPU, etc. I'm planning to get as much mileage out of it as I can. I'm sure a Studio would be more performant, but the Thunderbolt to PCIe chassis are not cheap.
I hate how I can't buy an new apple silicon with upgradeable RAM or SSD. Is there a legit reason why they couldn't make these things upgradeable at all even on a studio machine? 4TB is the smallest SSD I ever want in a new machine, but buying one from Apple is stupidly expensive. Back in the intel days, I'd buy a macbook pro, for example, with less ram and a smaller SSD than the max available and then upgrade to much cheaper aftermarket parts a few years later when prices dropped.
I'm still not going to use windows or linux. Don't want to be an IT guy on the side just to keep linux machines working. This may not be obvious to some unless you try to use printers and scanners that are more than 5 years old and what them to be on the network. And, you don't install virtualization tools like vmware that require compiling and loading kernel drivers which ends up being incompatible with new OS releases...etc.
Windows is just too much of a painful acceptance of mediocrity and apathy in product design for me.
They've been trying to kill the Mac Pro for over a decade. I wonder how long before they backtrack again? It seems like they should at least have a migration path for users who needed the expansion cards the Mac Pro supported. Pushing them to the PC seems pretty bad.
Apple's new "Pro" definition seems more like "Prosumer".
The form-factor always felt like a weird fit for Apple Silicon. With the Intel boxes it was understandable; you want a few liters of free space for a couple AMD cards or some transcode hardware. The system was designed to be expandable, and the Mac Pro was the apex of Apple's commitment to that philosophy after bungling the trashcan Mac Pro.
None of the Apple Silicon hardware can seemingly justify this form factor, though. The memory isn't serviceable, PCIe devices aren't really supported, the PSU doesn't need much space, and the cooling can be handled with mobile-tier hardware. Apple's migration path is "my way or the highway" for Mac Pro owners.
Their justification for the form factor, when it was released, was that pro users need various PCI cards to interface with some of their equipment, and this would allow them to do that.
It seemed like the guts of the Mac Pro were essentially shoved inside of a box and stuck in the corner of the tower. It would seem like they could decouple it and sell a box that pro users could load cards into (like other companies do for eGPUs). It wouldn’t feel like a very Apple-like setup, but it would function and allow Apple to focus where they want to focus without simply leaving those users behind.
I suppose the other option would be to dispense with the smoke and mirrors and let people slot a Mac Studio right into the Mac Pro tower, so it could be upgraded independently of the tower.
The alternative is people leave the platform or end up with a bunch of Thunderbolt spaghetti. Neither of which seem ideal.
I suspect we'll start seeing higher-spec Mac Studio options.
One of those with an M* Ultra, and some sort of Thunderbolt storage expansion would probably cover most of the Pro's use cases. And Apple probably doesn't want to deal with anything more exotic than those.
I have three of the trash can ones. They are absolute pieces of art, as useless as they are computationally these days (energy-to-performance wise at least). I will never sell nor give them away.
With the popularity of mac mini (and macbooks for that matter) for doing ML/AI work, I would have thought Apple could make a Mac Pro that could make for a good workstation for doing in-house ML/AI stuff.
I bought a GPU maybe a decade ago for this, and it's not worth the hassle (for me at least), but a nice out-of-the box solution, I would pay for.
The problem is that the M1 chips foretold the doom of the Mac Pro unless they could figure out some way to do something that you couldn't do with a Mac Studio - thunderbolt is so good that it's hard to justify anything else.
If they had done more with NUMA in the M series maybe you could have a Mac Pro with M5 Ultras that can take a number of M5 "daughter cards" that do something useful.
Reading comments, I don’t think people are being completely fair here. For Intel and AMD to approach what Apple has accomplished they’re making many of the same compromises with Panther Lake and Ryzen AI Max. Apple chose to put disk controllers on their SoP rather than having them on the storage module. This shaves a tiny bit of latency. Worth it? No idea. I’m shit at hardware design.
As for not having a Pro or otherwise expandable system? It’s shit. They make several variations of their chips, and I don’t think it would hurt them to make an SoP for a socket, put a giant cooling system in it, and give it 10 or 12 PCIe slots. As for what would go in those slots? Make this beast rack mountable and people would toss better network cards, sound/video output or capture, storage controllers, and all kinds of other things in there. A key here would be to not charge so much just because they can. Make the price reasonable.
They have tried variations of this since time immemorial (we can argue about "price reasonablé") but there's just not much you can do with it that you can't do much cheaper or simpler in other ways.
The Xserve has been dead for 15 years now, and it was never tremendously amazing (though it was nice kit).
Apple apparently has some sort of "in-house" xserve-like thing they don't sell; but turning that into a product would likely be more useful than a Mac Pro, unless they add NUMA or some other way of allowing an M5 to access racks and racks of DIMMs.
Mac Studio waits for the Ultra chips to ship, which are always last in a generation. Perhaps the M5's chiplet architecture will help them move faster there.
The 2013 trash can was the end of the Mac Pro. It was never the same after that. The 2012 and earlier Mac Pros were awesome. I had a 2010 model. Here's what I loved:
• Multiple hard drive bays for easy swapping of disks, with a side panel that the user could open and close
• Expandable RAM
• Lots of ports, including audio
• The tower took up no desktop space
• It was relatively affordable, starting at $2500. Many software developers had one. (The 2019 and later Mac Pros were insanely expensive, starting at $6000.)
The Mac Studio is affordable, but it lacks those other features. It has more ports than other Macs but fewer in number and kind than the old Mac Pro, because the Mac Studio is a pointlessly small desktop instead of floor tower.
That's when they stopped designing computers for the pro market and started selling mid-century Danish furniture that can also edit videos.
I knew it was all over when third party companies had to develop the necessarily-awkward rack mount kits for those contraptions. If Apple actually cared about or understood their pro customers, they would have built a first party solution for their needs. Like sell an actual rack-mount computer again—the horror!
Instead, an editing suite got what looked like my bathroom wastebasket.
When it was introduced, Apple said the trash can was a revolution in cooling design.
Then they said they couldn't upgrade the components because of heat. Everyone knows that wasn't true.
By the time Apple said they had issues with it in 2017, AMD were offering 14nm GCN4 and 5 graphics (Polaris and Vega) compared to the 28nm GCN1 graphics in the FirePro range. Intel had moved from Ivy Bridge to Skylake for Xeons. And if they wanted to be really bold (doubtful, as the move to ARM was coming) then the 1st gen Epyc was on the market too.
Moore's Law didn't stop applying for 6 years. They had options and chose to abandon their flagship product (and most loyal customers) instead.
The biggest issue was actually that the Mac Pro was designed specifically for dual GPUs- in the era of SLI this made some sense, but once that technology was abandoned it was a technological dead-end.
If you take one apart you'll see why, it's not the case that you could have ever swapped around the components to make it dual-CPU instead; it really was "dual GPU or bust".
Somewhat ironically, in todays ML ecosystem, that architecture would probably do great. Though I doubt it could possibly do better than what the M-series is doing by itself using unified memory.
I'll admit that while I've used the trash can but never taken it apart myself. But I can't imagine it would have been impossible to throw 2x Polaris 10 GPUs on the daughterboards in place of the FirePros.
For what is essentially a dead-end technology, I'm somewhat doubtful people would have bought it (since the second GPU is going to be idle and add to the cost massively).
the CPU being upgraded would have been much easier though I think.
Aside from the GPU mess, the 2013 was a nice machine, basically a proto-Mac Studio. Aside from software, the only thing that pushed me off my D300/64GB/12-core as an everyday desktop + front-end machine is the fact that there's no economically sensible way to get 4K video at 120 Hz given that an eGPU enclosure + a decent AMD GPU would cost as much as a Mac mini, so I'm slumming it in Windows for a few months until the smoke clears from the next Mac Studio announcement.
At which point I'll decide whether to replace my Mac Pro with a Mac Studio or a Linux workstation; honestly, I'm about 60/40 leaning towards Linux at this point, in which case I'd also buy a lower-end Mac, probably a MacBook Air.
I'm in the Linux desktop / Mac laptop camp, and it works well for me. Prevents me getting too tied up in any one ecosystem so that I can jump ship if Apple start releasing duds again.
If I remember correctly, the maximum configuration was something like $35k back in the day. I wonder what those people feel like now. On the other hand, if they have $35k to burn, probably they don't even think about it.
Siracusa—probably best known here for doing fabulous OS X reviews for Ars—is a co-host of ATP. He is also known is such circles for having Mac Pros, and using them for a long time (sometimes by choice, sometimes by circumstance). He thinks Apple should make a Mac Pro, not necessarily because it's a big seller, but because he thinks Apple should make a "best computer," much in the same way car companies might make a car that will never sell but pushes engineers, etc.
Ages ago, when new Mac hardware came out, I'd amuse myself by putting together an "ultimate Mac workstation" in the configurator --- once upon a time, one could hit 6 figures pretty easily --- these days, well I panic bought a duplicate computer because I was worried a chipped/cracked display was going to make it unusable (turns out a screen protector has worked thus far).
I agree with the reasoning, and would like to see Apple continue to make aspirational hardware, but maybe the mainstream stuff is good enough?
Even Siracusa admits that - he's found it hard to articulate what a true "Mac Pro" would do that you can't do with other things.
Back in the heyday of the $100k Mac Pro you could certainly imagine it doing things that wouldn't be easily done by anything under $50k, and it would look good doing it.
As someone who came from the SGI O2/Octane era when high-end workstations were compact, distinctive, and sexy, I’ve never really understood the allure of the Mac Pro, with the exception of the 2013 Mac Pro tube, which I owned (small footprint, quiet, and powerful).
For me, aesthetics and size are important. That workstation on your desk should justify its presence, not just exist as some hulking box.
When Apple released the Mac Studio, it made perfect sense from a form-factor point-of-view. The internal expansion slots in the M2 Mac Pro didn't make any sense. It was like a bag of potato chips - mostly air. And far too big and ugly to be part of my work area! I'm surprised that Apple didn't discontinue it sooner.
As someone who worked on the M2 Mac Pro and has a real soft spot for it, I get it. It’s horrendously expensive and doesn’t offer much benefit over a Mac Studio and a thunderbolt pci chassis. My personal dream is that vms would support pci pass through and so you can just spin up a Linux vm and let it drive the gpus. But at that point, why are you buying a Mac?
Opinions are my own obvs.
> My personal dream is that vms would support pci pass through and so you can just spin up a Linux vm and let it drive the gpus.
SR-IOV is just that? and is well supported by both Windows and Linux.
SR-IOV and VFIO passthrough are different things. SR-IOV partitions a PCIe device across multiple VMs simultaneously (common for NICs and NVMe). VFIO passthrough gives one VM exclusive ownership of a physical device. For GPU compute you almost always want full passthrough, not SR-IOV partitioning.
The harder problem on Apple Silicon is that the M2 Ultra's GPU is integrated into the SoC -- it's not a discrete PCIe device you can isolate with an IOMMU group. Apple's Virtualization framework doesn't expose VFIO-equivalent hooks, so even if you add a discrete AMD Radeon to the Mac Pro's PCIe slots, there's no supported path to pass it through to a Linux guest right now.
On Intel Macs this actually worked via VFIO with the right IOMMU config. Apple Silicon VMs can do metal translation layers but that's not the same as bare-metal GPU access. It's a real limitation and I doubt Apple will prioritize solving it since it would undercut the "just use macOS" pitch.
> Opinions are my own obvs.
Whose else would they be?
> as someone who worked on the m2 mac pro
They're trying to make it very clear they're not speaking on behalf of Apple Inc, despite having worked (or working) there.
Big companies like to give employees some minimal "media training", which mostly amounts to "do not speak for the company, do not say anything that might even slightly sound like you're speaking for the company".
Maybe he was trying to say he isn’t a spokesman for anyone else :-)
> aesthetics and size are important
It's dumb from a practical perspective. But I keep hoping they'll vertically compress their trashcan design so it looks like their Cupertino headquarters.
> That workstation on your desk should
Under your desk, right? Right?!
i wish i'd never traded in my 2016 mac pro (aluminum polished tube) as it was beefy, it was silent, clever thermo design (like the powerpc cube 20 years earlier or so), and i'd upgraded the living crap out of it for cheap.
I'm surprised they even tried selling an Apple Silicon Mac Pro - I expected that product to die the moment they announced the transition. Everything that makes Apple Silicon great also makes it garbage for high-performance workstations.
The allure of the Mac Pro is that you could dodge the Apple Tax by loading it up with RAM and compute accelerators Apple couldn't mark up. Well, Apple Silicon works against all of that. The hardware fabric and PCIe controller specifically prohibit mapping PCIe device memory as memory[0], which means no GPU driver ever will work with it. Not even in Asahi Linux. And the RAM is soldered in for performance. An Ultra class chip has like 16 memory channels, which even in a 1-DIMM per channel routing would have trace lengths long enough to bottleneck operating frequency.
The only thing the socketed RAM Mac Pros could legitimately do that wasn't a way to circumvent Apple's pricing structure was take terabytes of memory - something that requires special memory types that Apple's memory controller IP likely does not support. Intel put in the engineering for it in Xeon and Apple got it for free before jumping ship.
Even then, all of this has gone completely backwards. Commodity DRAM is insanely expensive now and Apple's royalty-bearing RAM prices are actually reasonable in comparison. So there's no benefit to modularity anymore. Actually, it's a detriment, because price-discovery-enforcing scalpers can rip RAM out of perfectly working computers and resell the RAM. It's way harder to scalp RAM that's soldered on the board.
[0] In violation of ARM spec, even!
For AI Apple internally has rack mounted M chips but they wont sell them to the public.
Apple really dropped the ball here. They had every ability to make something competitive with Nvidia for AI training as well as inference, by selling high end multi GPU Mac Pro workstations as well as servers, but for some reason chose not to. They had the infrastructure and custom SoCs and everything. What a waste.
It really could have been a bigger market for them than even the iPhone.
Just about everybody who isn't Nvidia dropped the ball, bigtime.
Intel should have shipped their GPUs with much more VRAM from day one. If they had done this, they'd have carved out a massive niche and much more market share, and it would have been trivially simple to do.
AMD should have improved their tools and software, etc.
Apple should have done as you say.
Google had nigh on a decade to boost TPU production, and they're still somehow behind the curve.
Such a lack of vision. And thus Nvidia is, now quite durably, the most valuable company in the world. Imagine telling that to a time traveler from 2018.
I think for AMD, they were focused on competing against Intel. Remember AMD was almost bankrupt about 15 years ago because of competing against Intel. But the very first GPU use for AI was actually with an ATI/AMD GPU, not an Nvidia one. Everyone thinks Nvidia kicked off the GPU AI craze when Ilya Sutskever cleaned up on AlexNet with an Nvidia GPU back in 2012, or when Andrew Ng and team at Stanford published their "Large Scale Deep Unsupervised Learning using Graphics Processors" in 2009, but in 2004, a couple of Korean researchers were the first to implement neural networks on a GPU, using ATI Radeons: https://www.sciencedirect.com/science/article/abs/pii/S00313...
And as of now I do believe AMD is in the second strongest position in the datacenter space after Nvidia, ahead of even Google.
> And thus Nvidia is, now quite durably, the most valuable company in the world.
Nvidia is the most valuable company in the world right up until the AI bubble pops. Which, while it's hard to nail down when, is going to happen. I wouldn't call their position durable at all.
Don’t mistake stock market performance for revenue. NVIDIA makes ~200B annually, same as what Apple makes from iPhones. It’s a big market but GPUs aren’t just AI.
I'm purely talking in terms of revenue. There's a huge demand for AI systems from personal workstations to datacenter servers, and Apple was one of the few companies in the world in a position to build complete systems for it.
But for some reason Apple thought the sound recording engineer or the video editor market was more important... like, WTF dude? Have some vision at least!
Some people at Apple see it. That’s why they added matmul to M5 GPU and keep mentioning LMStudio in their marketing.
280b and growing 70% YoY.
Nah, Apple made the right choice. Nobody except a niche market of hobbyists is interested in running tiny quantized models.
> something competitive with Nvidia for AI training
Apple is counting on something else: model shrink. Every one is now looking at "how do we make these smaller".
At some point a beefy Mac Studio and the "right sized" model is going to be what people want. Apple dumped a 4 pack of them in the hands of a lot of tech influencers a few months back and they were fairly interesting (expensive tho).
> Apple is counting on something else: model shrink
The most powerful AI interactions I've had involved giving a model a task and then fucking off. At that point, I don't actually care if it takes 5 minutes or an hour. I've cued up a list of background tasks it can work on, and that I can circle back to when I have time. In that context, smaller isn't even the virtue at hand–user patience is. Having a machine that works on my bullshit questions and modelling projects at one tenth the speed of a datacentre could still work out to being a good deal even before considering the privacy and lock-in problems.
> At some point a beefy Mac Studio and the "right sized" model is going to be what people want.
It's pretty clear that this isn't going to happen any time soon, if ever. You can't shrink the models without destroying their coherence, and this is a consistently robust observation across the board.
I don’t think it’s about literally shrinking the models via quantization, but rather training smaller/more efficient models from scratch
Smaller models have gotten much more powerful the last 2 years. Qwen 3.5 is one example of this. The cost/compute requirements of running the same level intelligence is going down
Cheaper than what you’d expect though. You could get a nice setup for $20-40k 6mo ago. As far as enterprise investments go, that’s a rounding error.
Not all enterprises are the same, I imagine many companies have different departments working with local optimums, so someone who could benefit from it to get more productivity might not have access to it because the department that is doing hardware acquisition is being measured in isolation.
Drop that down to 5k, and make it useful.
Give every iPhone family a in house Siri that will deal with canceling services and pursuing refunds.
Your customer screw up results in your site getting an agent drive DDOS on its CS department till you give in.
Siri: "Hey User, here's your daily update, I see you haven't been to the gym, would you like me to harass their customer service department till they let you out of their onerous contract?"
Nothing is a bigger market than the iPhone, let alone expensive niche machines.
The Ultra variants of the M series chips had previously consisted of two of the Max chips bonded together.
The M5 generation Pro and Max chips have moved to a chiplet based architecture, with all the CPU cores on one chiplet, and all the GPU cores on another.
https://www.wikipedia.org/wiki/Apple_M5
So what will the M5 Ultra look like?
If you integrate two CPU chiplets and two GPU chiplets, you're looking at 36 CPU cores, 80 GPU cores, and 1228 GB/s of memory bandwidth.
The only thing that is missing from the current Mac line up is a one rack unit machine.
While the trash can generation was somewhat present and around, I don't think I ever saw a cheese grater in the flesh. Did it have any users? Were there any actual useful expansion cards? Did anybody continue buying this at all, after it didn't get the M3 Ultra bump, that the Mac Studio got last year?
I just replaced a 2009 MacPro
It had many hardware upgrades over the years - upgraded CPUs, 128GB RAM, 4TB NVME storage, a modern AMD GPU, USB3/c, thunderbolt, etc
The only reason it got replaced is because it became too much of a PITA to keep modern OSX running on it (via OCLP)
Replaced with an M4 Max Mac Studio, which is a nice and faster machine but with no ability to upgrade anything and much worse hardware resale value on M-series I'll have to replace it in 2-3 years
I'm a former 4,1 user, myself — replaced with an M2Pro mini Jan 2023 (finally retired fully 2025).
Absolutely recommend you purchase the 4-bay Terramaster external enclosure — gives you four SATA slots that are hot-swappable (unlike MacPro's). 10gbps via USB-C.
The cheese grater mac pros were very popular, in that people got them and continued to use them.
The most notable feature was that there were mac-specific graphics cards, and you could also run PC graphics cards (without a nice boot screen). They had a 1.4kw power supply I believe, and there was extra pcie power for higher-end graphics cards. You could upgrade the memory, add up to 6 or more sata hard disks (2 in dvd slot). You could run windows, dual booting if you wanted and apple supported the drivers.
The 2013 was kind of a joke. small and quiet, but expansion was minimal.
2019 looked beefy, but the expansion was more like a cash register for apple, not really democratic. There were 3rd party sata hard disk solutions,
the 2023 model was basically a joke. I think maybe the pcie slots were ok for nvme cards, not a lot else (unless apple made it).
nowadays an apple computer is more like an iphone - apple would prefer if everything was welded shut.
My first non-Linux PC was a cheese grater, way overkill for my needs but served me well for many years.
Not surprising, as the market has broadly moved on from add-in cards in favor of smaller form factors and external devices, absent some notable holdouts in specific verticals.
Gonna miss it, though. If they had reduced the add-in card slots to something more reasonable, lowered the entry price, and given us multi-socket options for the CPU (2x M# Ultras? 4x?), it could have been an interesting HPC or server box - though they’ve long since moved away from that in software land, so that was always but a fantasy.
At least the Mac Studio and Minis are cute little boxes.
This makes sense, for that kind of money you could always build a beastly workstation in a real ATX case with standard components. Install Linux and the Mac looks like an expensive toy in comparison.
Still rocking a 2019 (Intel) Mac Pro here, all slots filled with various Pro Tools and UAD DSP cards, SSD, GPU, etc. I'm planning to get as much mileage out of it as I can. I'm sure a Studio would be more performant, but the Thunderbolt to PCIe chassis are not cheap.
I feel like Apple is going back to the days of toaster "appliance" Macs. No slots, no upgrades, just buy a new one in 3-7 years.
I hate how I can't buy an new apple silicon with upgradeable RAM or SSD. Is there a legit reason why they couldn't make these things upgradeable at all even on a studio machine? 4TB is the smallest SSD I ever want in a new machine, but buying one from Apple is stupidly expensive. Back in the intel days, I'd buy a macbook pro, for example, with less ram and a smaller SSD than the max available and then upgrade to much cheaper aftermarket parts a few years later when prices dropped.
I'm still not going to use windows or linux. Don't want to be an IT guy on the side just to keep linux machines working. This may not be obvious to some unless you try to use printers and scanners that are more than 5 years old and what them to be on the network. And, you don't install virtualization tools like vmware that require compiling and loading kernel drivers which ends up being incompatible with new OS releases...etc.
Windows is just too much of a painful acceptance of mediocrity and apathy in product design for me.
I kinda would have loved a new Mac Pro, same case, but just stick 4 Mac Studios in there and connect them all via MLX.
Would be a killer local AI setup...for $40k.
They've been trying to kill the Mac Pro for over a decade. I wonder how long before they backtrack again? It seems like they should at least have a migration path for users who needed the expansion cards the Mac Pro supported. Pushing them to the PC seems pretty bad.
Apple's new "Pro" definition seems more like "Prosumer".
The migration path is Thunderbolt PCIe enclosures (basically eGPU enclosures but you don't have to use a GPU).
> but you don't have to use a GPU
That's a cute way of saying that GPUs aren't supported.
The form-factor always felt like a weird fit for Apple Silicon. With the Intel boxes it was understandable; you want a few liters of free space for a couple AMD cards or some transcode hardware. The system was designed to be expandable, and the Mac Pro was the apex of Apple's commitment to that philosophy after bungling the trashcan Mac Pro.
None of the Apple Silicon hardware can seemingly justify this form factor, though. The memory isn't serviceable, PCIe devices aren't really supported, the PSU doesn't need much space, and the cooling can be handled with mobile-tier hardware. Apple's migration path is "my way or the highway" for Mac Pro owners.
Their justification for the form factor, when it was released, was that pro users need various PCI cards to interface with some of their equipment, and this would allow them to do that.
It seemed like the guts of the Mac Pro were essentially shoved inside of a box and stuck in the corner of the tower. It would seem like they could decouple it and sell a box that pro users could load cards into (like other companies do for eGPUs). It wouldn’t feel like a very Apple-like setup, but it would function and allow Apple to focus where they want to focus without simply leaving those users behind.
I suppose the other option would be to dispense with the smoke and mirrors and let people slot a Mac Studio right into the Mac Pro tower, so it could be upgraded independently of the tower.
The alternative is people leave the platform or end up with a bunch of Thunderbolt spaghetti. Neither of which seem ideal.
It was always strange the Apple Silicon kept the 1.3kw power supply which was massive overkill.
I always hoped we’d get a consumer version of what they have internally - 10 or 20 or more Apple Silicon chips for 1000 cores or so.
A bunch of the Mac Pro decisions seem to have been more driven by "we have a warehouse of these parts" than "this is what the system needs".
A lot of Apples offerings feel a bit like that actually.
To be expected when lord of the supply chain Tim Cook is running the show.
I suspect we'll start seeing higher-spec Mac Studio options.
One of those with an M* Ultra, and some sort of Thunderbolt storage expansion would probably cover most of the Pro's use cases. And Apple probably doesn't want to deal with anything more exotic than those.
Can't you already stack 4 Studios? Building a $1,000 case to hold them makes some sort of Apple sense.
https://www.youtube.com/watch?v=x4_RsUxRjKU or something
I have three of the trash can ones. They are absolute pieces of art, as useless as they are computationally these days (energy-to-performance wise at least). I will never sell nor give them away.
With the popularity of mac mini (and macbooks for that matter) for doing ML/AI work, I would have thought Apple could make a Mac Pro that could make for a good workstation for doing in-house ML/AI stuff.
I bought a GPU maybe a decade ago for this, and it's not worth the hassle (for me at least), but a nice out-of-the box solution, I would pay for.
The problem is that the M1 chips foretold the doom of the Mac Pro unless they could figure out some way to do something that you couldn't do with a Mac Studio - thunderbolt is so good that it's hard to justify anything else.
If they had done more with NUMA in the M series maybe you could have a Mac Pro with M5 Ultras that can take a number of M5 "daughter cards" that do something useful.
I’m much less concerned about the Mac Pro and more concerned we won’t see an XDR Ultra Display to replace the 32”.
They did come out with a new XDR. It’s just not 32”.
Reading comments, I don’t think people are being completely fair here. For Intel and AMD to approach what Apple has accomplished they’re making many of the same compromises with Panther Lake and Ryzen AI Max. Apple chose to put disk controllers on their SoP rather than having them on the storage module. This shaves a tiny bit of latency. Worth it? No idea. I’m shit at hardware design.
As for not having a Pro or otherwise expandable system? It’s shit. They make several variations of their chips, and I don’t think it would hurt them to make an SoP for a socket, put a giant cooling system in it, and give it 10 or 12 PCIe slots. As for what would go in those slots? Make this beast rack mountable and people would toss better network cards, sound/video output or capture, storage controllers, and all kinds of other things in there. A key here would be to not charge so much just because they can. Make the price reasonable.
They have tried variations of this since time immemorial (we can argue about "price reasonablé") but there's just not much you can do with it that you can't do much cheaper or simpler in other ways.
The Xserve has been dead for 15 years now, and it was never tremendously amazing (though it was nice kit).
Apple apparently has some sort of "in-house" xserve-like thing they don't sell; but turning that into a product would likely be more useful than a Mac Pro, unless they add NUMA or some other way of allowing an M5 to access racks and racks of DIMMs.
WSJ recently did a thing about it but details were rather light
Honestly the Mac Studio is the new Mac Pro, this makes more sense to me.
but even that one looks kinda outdated when looking at latest M5 Max laptops.
Mac Studio waits for the Ultra chips to ship, which are always last in a generation. Perhaps the M5's chiplet architecture will help them move faster there.
The price to value for the Mac Pro kept going down. Mac Studio made more sense.
Another company would have killed it a long time ago. It lasted this long because it’s Apple.
The 2013 trash can was the end of the Mac Pro. It was never the same after that. The 2012 and earlier Mac Pros were awesome. I had a 2010 model. Here's what I loved:
• Multiple hard drive bays for easy swapping of disks, with a side panel that the user could open and close
• Expandable RAM
• Lots of ports, including audio
• The tower took up no desktop space
• It was relatively affordable, starting at $2500. Many software developers had one. (The 2019 and later Mac Pros were insanely expensive, starting at $6000.)
The Mac Studio is affordable, but it lacks those other features. It has more ports than other Macs but fewer in number and kind than the old Mac Pro, because the Mac Studio is a pointlessly small desktop instead of floor tower.
That's when they stopped designing computers for the pro market and started selling mid-century Danish furniture that can also edit videos.
I knew it was all over when third party companies had to develop the necessarily-awkward rack mount kits for those contraptions. If Apple actually cared about or understood their pro customers, they would have built a first party solution for their needs. Like sell an actual rack-mount computer again—the horror!
Instead, an editing suite got what looked like my bathroom wastebasket.
When it was introduced, Apple said the trash can was a revolution in cooling design.
Then they said they couldn't upgrade the components because of heat. Everyone knows that wasn't true.
By the time Apple said they had issues with it in 2017, AMD were offering 14nm GCN4 and 5 graphics (Polaris and Vega) compared to the 28nm GCN1 graphics in the FirePro range. Intel had moved from Ivy Bridge to Skylake for Xeons. And if they wanted to be really bold (doubtful, as the move to ARM was coming) then the 1st gen Epyc was on the market too.
Moore's Law didn't stop applying for 6 years. They had options and chose to abandon their flagship product (and most loyal customers) instead.
The biggest issue was actually that the Mac Pro was designed specifically for dual GPUs- in the era of SLI this made some sense, but once that technology was abandoned it was a technological dead-end.
If you take one apart you'll see why, it's not the case that you could have ever swapped around the components to make it dual-CPU instead; it really was "dual GPU or bust".
Somewhat ironically, in todays ML ecosystem, that architecture would probably do great. Though I doubt it could possibly do better than what the M-series is doing by itself using unified memory.
I'll admit that while I've used the trash can but never taken it apart myself. But I can't imagine it would have been impossible to throw 2x Polaris 10 GPUs on the daughterboards in place of the FirePros.
I think on a technical level you're right, but you need to run two of them and they'd need a custom design like so:
https://i.ebayimg.com/images/g/RQIAAOSwxKFoTHe3/s-l1200.jpg
For what is essentially a dead-end technology, I'm somewhat doubtful people would have bought it (since the second GPU is going to be idle and add to the cost massively).
the CPU being upgraded would have been much easier though I think.
Aside from the GPU mess, the 2013 was a nice machine, basically a proto-Mac Studio. Aside from software, the only thing that pushed me off my D300/64GB/12-core as an everyday desktop + front-end machine is the fact that there's no economically sensible way to get 4K video at 120 Hz given that an eGPU enclosure + a decent AMD GPU would cost as much as a Mac mini, so I'm slumming it in Windows for a few months until the smoke clears from the next Mac Studio announcement.
At which point I'll decide whether to replace my Mac Pro with a Mac Studio or a Linux workstation; honestly, I'm about 60/40 leaning towards Linux at this point, in which case I'd also buy a lower-end Mac, probably a MacBook Air.
I'm in the Linux desktop / Mac laptop camp, and it works well for me. Prevents me getting too tied up in any one ecosystem so that I can jump ship if Apple start releasing duds again.
The studio is also like 5x as fast as those machines.
What's your point? Of course processors have gotten a lot faster between 2012 and 2025.
I was talking about the form factor of the machine.
I mean for the cost of the Mac Pro wheels you can get a Macbook Neo these days!
In other news, the Mac Pro Wheels Kit was also discontinued.
https://www.macrumors.com/2026/03/26/mac-pro-wheels-kit-disc...
If I remember correctly, the maximum configuration was something like $35k back in the day. I wonder what those people feel like now. On the other hand, if they have $35k to burn, probably they don't even think about it.
Now everyone that needs classical workstations can finally move on into Linux or Windows workloads.
Believe t-shirts at WWDC were not enough.
Thus the workstation market joins OS X Server.
For those who don't know what the t-shirt reference is, it's a creation by John Siracusa/The Accidental Tech Podcast: <https://cottonbureau.com/p/4RUVDA/shirt/mac-pro-believe-dark>.
And I still don't get it.
Siracusa—probably best known here for doing fabulous OS X reviews for Ars—is a co-host of ATP. He is also known is such circles for having Mac Pros, and using them for a long time (sometimes by choice, sometimes by circumstance). He thinks Apple should make a Mac Pro, not necessarily because it's a big seller, but because he thinks Apple should make a "best computer," much in the same way car companies might make a car that will never sell but pushes engineers, etc.
They made a shirt. It was fun.
Ages ago, when new Mac hardware came out, I'd amuse myself by putting together an "ultimate Mac workstation" in the configurator --- once upon a time, one could hit 6 figures pretty easily --- these days, well I panic bought a duplicate computer because I was worried a chipped/cracked display was going to make it unusable (turns out a screen protector has worked thus far).
I agree with the reasoning, and would like to see Apple continue to make aspirational hardware, but maybe the mainstream stuff is good enough?
> maybe the mainstream stuff is good enough?
Even Siracusa admits that - he's found it hard to articulate what a true "Mac Pro" would do that you can't do with other things.
Back in the heyday of the $100k Mac Pro you could certainly imagine it doing things that wouldn't be easily done by anything under $50k, and it would look good doing it.
Apple still sells a workstation-type machine: the Mac Studio.
What is this, a workstation for ants?
It's a pizza box, for a 6" pizza.