Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
The one use case that require massive compute is genai and there we are still talking about the need for something like a 4xultra for anything approaching what you get from frontier model providers.
Nevertheless, Mac Studios and Minis have found a niche in local AI where the Apple Silicon unified RAM architecture - with the same RAM being shared by the CPU, GPU and NPU actually has an advantage over PCIe systems that separate system memory and VRAM. Meanwhile, NVIDIA and AMD are both making AI workstations that are more like the Mac Studio/Mini than a conventional tower.


Mac Studio has to be the most boring looking product they have produced. I mean seriously, the 2013 mac pro with AS components would have been quite sick
Have to disagree there.

Form should follow function. In the case of the 2013 Mini Apple had a thermal design consisting of three heat sources on vertical PCBs arranged around a roughly triangular central "chimney" heatsink, which justified the cylindrical shape.

Apple Silicon has a single primary heat source (and probably doesn't generate as much heat) so there's no good reason to make it circular - which would just add cost, complication (with things like port placement) and probably enlarge the footprint. Cuboid may be boring but it's easy to stack and mount, and its low eough to sit in the "dead space" under the display screen.

...and added to that, the 2013 thermal design was a disaster which didn't work too well to start with and tied the design to a particular chipset and helped make it a one-and-done product. One wonders whether it was "form-follows-function" or "cool form justifies dodgy function decisions".

Maybe they could find a less boring design for the Studio - the 'classic Mac Mini' footprint was initially determined by the need to hold an optical drive - but please don't copy the trashcan. I'd rather they looked at real issues like easily cleanable filters and more easily user-replaceable SSDs...
 
With RAM prices, $60,000 with 512GB or $120,000 with 1TB. Haha probably not far off.

It won't be that bad. But I'm bracing myself for sticker shock. On the refurb units, it costs $8,840 to go from 96GB to 512GB. So, I'm guesstimating $10k will be the additional price to go from 96GB to 512GB. And the base price will probably increase to boot. It's gonna feel bad.
 
Still very sad they left the high-end workstation space. For the good of the platform's health there should be a diversity of offerings across most of the needs that people and companies have for computers.
 
  • Like
Reactions: wyliej
I still have my 2008 MacPro tower and it still runs great. It was my main machine until 2022, believe it or not.

Best computer ever. Hands down.

And IMHO I think it's a mistake for Apple to abandon it. The ability to have multiple internal HDs, tons of IO ports, and owner-upgraded RAM etc, is still very desirable.
 
  • Like
Reactions: wyliej
Problem is that Apple mismanaged the Mac Pro. There was a lot of room in the market for it, but Apple didn't give it enough love and attention.

The 2019 model especially was so close to perfect in so many ways. I guess it was always too PC-like for their tastes. That model was such a tease because after so many years of messing around with it, it seemed they finally understood.

And then they did what they did to it, and it was just...gone...
 
  • Like
Reactions: Schismz
Very powerful Mac. Mac Studio will definitely become more powerful with future versions. Cannot imagine how much more expensive it will become.
It is way more powerful already. My Mini M2 Pro essentially retired the fully upgraded 12-core Mac Pro. That M2 is so much faster in single core processing, it’s crazy.
 
The Cheesegraters were the best workstation Macs. We had many of them in a printing shop over the years. Easy to upgrade, easy to fix. Never touched the Trashcan MP, and not the 2019 because it was insanely expensive, and the 2022 because it just never made any sense.
 
The 2019 model especially was so close to perfect in so many ways. I guess it was always too PC-like for their tastes. That model was such a tease because after so many years of messing around with it, it seemed they finally understood.

And then they did what they did to it, and it was just...gone...
The Apple Silicon MP could have been modular, like using the size of the chassis to add extra Apple Silicon CPU/GPU modules, and having the ability to upgrade to newer generation modules.

The way they designed it was completely stupid with just pci-e slots that could accept a limited number of compatible cards.
 
I liked my Trashcan Mac: so quiet and unobtrusive.

I like my current Mac Pro (2019) but in can be very loud at times. I also do not enjoy having to unplug every single cable to get to the inside. The old case design where the side panel just popped off was much better.
 
Nevertheless, Mac Studios and Minis have found a niche in local AI where the Apple Silicon unified RAM architecture - with the same RAM being shared by the CPU, GPU and NPU actually has an advantage over PCIe systems that separate system memory and VRAM. Meanwhile, NVIDIA and AMD are both making AI workstations that are more like the Mac Studio/Mini than a conventional tower.



Have to disagree there.

Form should follow function. In the case of the 2013 Mini Apple had a thermal design consisting of three heat sources on vertical PCBs arranged around a roughly triangular central "chimney" heatsink, which justified the cylindrical shape.

Apple Silicon has a single primary heat source (and probably doesn't generate as much heat) so there's no good reason to make it circular - which would just add cost, complication (with things like port placement) and probably enlarge the footprint. Cuboid may be boring but it's easy to stack and mount, and its low eough to sit in the "dead space" under the display screen.

...and added to that, the 2013 thermal design was a disaster which didn't work too well to start with and tied the design to a particular chipset and helped make it a one-and-done product. One wonders whether it was "form-follows-function" or "cool form justifies dodgy function decisions".

Maybe they could find a less boring design for the Studio - the 'classic Mac Mini' footprint was initially determined by the need to hold an optical drive - but please don't copy the trashcan. I'd rather they looked at real issues like easily cleanable filters and more easily user-replaceable SSDs...

The caveats with the DGX Spark are the following:
  • memory size: only 128GB vs. up to 512GB on Mac. That is, that Mac will run bigger models
  • memory bandwidth: one-third that of an M3-Ultra. The Mac will be faster if you are streaming weights from memory constantly. Those fast NVIDIA GPUs don't give an advantage when they are sitting waiting for weights.
  • I/O: 4xUSB-C (10Gb/s) vs. 6xTB5 (80Gb/s) on Mac.
  • NVIDIA's version of Linux vs. Mac OS.
The high-level comparison of the two machines is this: An M3 Ultra is a high-end general-purpose desktop and wins on model capacity and, depending on what you are doing, speed. A DGX Spark wins on price and it gives you a first-class AI research/development environment (CUDA compatibility, NVIDIA AI software, etc.). You have to look at specs and your computing requirements to figure which is best for you.

Still, that <$5k price makes them quite attractive. I may wind up with one just for the development environment and CUDA compatibility.
 
I may turn on my iMacPro1,1 this weekend and pour one out in honor/remembrance. I'll have to pull it out of storage first.
 
The caveats with the DGX Spark are the following:
Sure - but rather than get into "which one would win in a fight" my point was really that NVIDIA are producing systems that have more in common with Apple Silicon Max/Ultra and the Mac Studio than they do with traditional x86/PCIe towers.

(I think one difference is that NVIDIA fuse a CPU die with a GPU die rather than two identical SOC dies - but we'll see what happens with the next Mx Ultra chip...)
 
  • Like
Reactions: vantelimus
The Apple Silicon MP could have been modular, like using the size of the chassis to add extra Apple Silicon CPU/GPU modules, and having the ability to upgrade to newer generation modules.

The way they designed it was completely stupid with just pci-e slots that could accept a limited number of compatible cards.

Yep. Especially after the 2019 model, that really seemed like they were overtly trying to kill it.

I don’t believe that Apple didn’t realize exactly what they were doing. It was literally the trash can Mac all over again, except this time it had TOO MUCH cooling.

They can’t possibly have missed that irony.
 
  • Like
Reactions: HiVolt
"That steady cadence of upgrades continued through 2012, a sharp contrast to the long droughts that would define the machine's later years."

The Mac Pro died when Steve died. It was always supposed to be an attainable halo product. Then Tim Cook's Apple removed the attainable element and mistook what the machine represented.

Sad ending to the Mac Pro story.
 
I got one. Sitting there not being used, but still boots up and is usable for simple stuff. Thinking of using its chassis for a bunch of Raspberry Pi's.
 
The high end trash can would get so hot that the GPU chips would unsolder from their circuit boards. OOPS!

I kept an 8 core 2010 Mac Pro running till 2023.
I think that I kept a Mac running for a little longer than that without issue.
1786129217758.png

Sorry for your loss, though.
 
  • Like
Reactions: ThomasJL
It’s getting to where a computer this size simply isn’t necessary. Especially with Apple Silicon. A Mac Studio or something a little bit larger larger with an Extreme chip would be beyond enough for something insanely powerful.
 
Just remember that Apple went from a 64 bit G5 power PC to a 32 bit bios Intel power Mac running 64 bit Xeon processors cutting off software updates at Mac OS X 10.7. DOSDUDE did provide a temporary workaround to get to El Capitan but it was never supported by Apple. 😵
 
The way they designed it was completely stupid with just pci-e slots that could accept a limited number of compatible cards.
I'm pretty sure that the 2023 Mac Pro was purely intended for users who had things like PCIe ProTools cards or specialist video digitisers, giving them a route to Apple Silicon and another 3 years to transition to Thunderbolt. It had better PCIe bandwidth than a Thunderbolt 3 PCIe dock. That's about it.

I don’t believe that Apple didn’t realize exactly what they were doing. It was literally the trash can Mac all over again, except this time it had TOO MUCH cooling.
The Mac Pro was killed by Apple Silicon. Apple Silicon is designed to use Thunderbolt as it's primary I/O and the Mx Pro/Max only has about 8 free PCIe lanes. The M3 Ultra has 24 spare PCIe lanes - as a side effect of being two M3 Max's fused together, adding 8 more plus another 8 from the 'spare' SSD interface. c.f. 64 on the Xeon W in the 2019 MP and probably 128+ on a modern workstation x86.

An Apple Silicon Mac with enough PCIe lanes and bandwidth to run multiple 16-lane GPU cards (or Apple Silicon 'compute units') isn't going to happen unless Apple designs a whole new die for the application. For such a tiny slice of the Mac market, that would be eye-wateringly expensive vs. the Ultra chips which doubled up on existing Mx Max die designs used in bigger-selling MacBook Pros and Studios.
 
  • Like
Reactions: CarAnalogy
I'm pretty sure that the 2023 Mac Pro was purely intended for users who had things like PCIe ProTools cards or specialist video digitisers, giving them a route to Apple Silicon and another 3 years to transition to Thunderbolt. It had better PCIe bandwidth than a Thunderbolt 3 PCIe dock. That's about it.


The Mac Pro was killed by Apple Silicon. Apple Silicon is designed to use Thunderbolt as it's primary I/O and the Mx Pro/Max only has about 8 free PCIe lanes. The M3 Ultra has 24 spare PCIe lanes - as a side effect of being two M3 Max's fused together, adding 8 more plus another 8 from the 'spare' SSD interface. c.f. 64 on the Xeon W in the 2019 MP and probably 128+ on a modern workstation x86.

An Apple Silicon Mac with enough PCIe lanes and bandwidth to run multiple 16-lane GPU cards (or Apple Silicon 'compute units') isn't going to happen unless Apple designs a whole new die for the application. For such a tiny slice of the Mac market, that would be eye-wateringly expensive vs. the Ultra chips which doubled up on existing Mx Max die designs used in bigger-selling MacBook Pros and Studios.

I suppose my old dream of the ultimate machine that could dual boot Mac and Windows and do anything will wither away like the floppy drive.

You make excellent points, of course. It hardly ever made economic sense. It just doesn't anymore.
 
The Mac Pro benefitted all Mac users. That's because for people who could only afford the cheapest budget-level Macs, those budget-level Macs would eventually get some of the Mac Pro's cutting-edge tech after a couple of years. The existence of the Mac Pro line allowed Apple to innovate without as much restriction for cost-cutting, since professional-level customers were often willing to spend a lot more than consumer-level and budget-level customers.
 
Have to disagree there.

Form should follow function. In the case of the 2013 Mini Apple had a thermal design consisting of three heat sources on vertical PCBs arranged around a roughly triangular central "chimney" heatsink, which justified the cylindrical shape.

Apple Silicon has a single primary heat source (and probably doesn't generate as much heat) so there's no good reason to make it circular - which would just add cost, complication (with things like port placement) and probably enlarge the footprint. Cuboid may be boring but it's easy to stack and mount, and its low eough to sit in the "dead space" under the display screen.

...and added to that, the 2013 thermal design was a disaster which didn't work too well to start with and tied the design to a particular chipset and helped make it a one-and-done product. One wonders whether it was "form-follows-function" or "cool form justifies dodgy function decisions".

Maybe they could find a less boring design for the Studio - the 'classic Mac Mini' footprint was initially determined by the need to hold an optical drive - but please don't copy the trashcan. I'd rather they looked at real issues like easily cleanable filters and more easily user-replaceable SSDs...
I have a mac studio sitting here on my desk and my biggest gripe by far is that you can't just upgrade the storage. As big as the stupid ugly metal brick is, it doesn't have removable m.2 slots. And I would LOVE removable m.2 slots.

We could go on and on about how people who buy a mac studio can afford to just trade it in for a new one with more storage or memory, but that involves downtime and swapping and costs. Or I could, ya know, flip the thing over, open a door, and add more storage?

As big as it is, it's a stupid design to not have easily upgradable storage.

The externals are a much much worse solution. Again, saying this as a person with a mac studio on their desk... and external SSDs connected to it...
 
  • Like
Reactions: opeter
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.