You can get the same 192GB in the Mac Studio. When Mac Pro was going to get an M2 Extreme (with up to 384GB) that may have made more sense. But I think Tim Cook just wanted to officially end the transition, particularly since he announced 3 years ago almost to the day that it would take “about 2 years.” Maybe the next Mac Pro will have an M3 or M4 Extreme with 384GB or even 768GB of unified RAM. Until then the Studio likely makes sense for more people.
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.
With M2 Ultra Mac Pro, Apple Completes Apple Silicon Transition and No Longer Sells Intel-Based Macs
- Thread starter MacRumors
- Start date
- Sort by reaction score
The most underwhelming product launched today IMO, apart from the PCI slots, it's no faster or more powerful than the Studio.....then again, I'm no expert and certainly not in the market for this type of product
Apple today introduced an Apple silicon version of the Mac Pro that uses the new M2 Ultra chip, and with that update, Apple's transition to Apple silicon is now complete. The first Apple silicon Mac came out in 2020, and three years later, every Mac is using Apple-designed chips.
The Mac Pro was the last Mac that was still using older Intel chip technology, and with the launch of the new M2 Ultra model, the Intel versions have been discontinued.
Apple may still be selling refurbished Intel Macs through its online store for refurbished devices, but none of its current product lineup is using Intel's chip technology.
The M2 Ultra chip is available in both the Mac Pro and the Mac Studio, both of which can be preordered today and will launch next week. The M2 Ultra Mac Pro is priced starting at $7,000, while the M2 Ultra Mac Studio is priced starting at $4,000.
Article Link: With M2 Ultra Mac Pro, Apple Completes Apple Silicon Transition and No Longer Sells Intel-Based Macs
My main desktop computer runs high sierra but my main desktop computer isn’t the one I use for email and internet etc. I have work to do.
I really do enjoy the popcorn nature of comments - and yeah the irony isn't lost open me.
No body knows where this machine will go, wasn't that long ago many were saying it won't have internal PCI it done be expandable.... but here we have a machine that does have slots and expansion but as yet I suspect all pieces of puzzle not in place. It really is quite amazing what can be released and for that matter gimped with firmware.
For 90% people the mini is all they really need, for the next 8-9% the studio in its forms and then the last 1% this device - its always been niche and always been high end - go back to days of IIfx, Quadras, 9600s, G5 Towers, MacPro Cheesegraters etc etc -they were always really really expensive and not targeted at any normal person.
I think we should be grateful its not totally sealed and is expandable and hopefully watch this space as things develop for those slots
No body knows where this machine will go, wasn't that long ago many were saying it won't have internal PCI it done be expandable.... but here we have a machine that does have slots and expansion but as yet I suspect all pieces of puzzle not in place. It really is quite amazing what can be released and for that matter gimped with firmware.
For 90% people the mini is all they really need, for the next 8-9% the studio in its forms and then the last 1% this device - its always been niche and always been high end - go back to days of IIfx, Quadras, 9600s, G5 Towers, MacPro Cheesegraters etc etc -they were always really really expensive and not targeted at any normal person.
I think we should be grateful its not totally sealed and is expandable and hopefully watch this space as things develop for those slots
My theory is the the "extreme" 4x Max chip didn't scale well.. 4x the cost.. and 2.5? x the performance of a max chip based on how poorly the M! ultra scaled. I am curious to see if the M2 Ultra scales way better than the M! ultra did. I couldn't tell from all the apple marketing speak wit they were just comparing the delta in the M2 Ultra to. The Asi Mac Pro is a huge failure.. a massive compromise to get something out the door when plan A, and possibly B failed. I would of rather seen an external GPU option for the Mac Studio than Pay see a $2000 mark up for a big empty Aluminum box to screw over Logic users.
Congratulations? They tried to create a real AS Mac Pro, but gave up and put it in the too hard basket. The new flagship Mac Pro is now a Mac Studio embarrassingly shovelled into a Mac Pro case. There is no way on earth that this is comparable to the Intel Mac Pro beyond the equally embarrassing, overpriced break-less wheels. Apple shouldn't be congratulated on this.... thing... they should be disgusted with themselves.
They already explain the reasoning in the keynote: "There are some pro users that depend on internal PCI expansion"
If it wasn't worth the cost for the market then Apple would have developed a new form factor. I suspect they might have decided to reuse it because they might try to develop a new AS architecture (meaning non-SOC), that can "grow into" the current design. Maybe they did learn a lesson after admitting the "thermal corner" issue with the trash can (?).
The other reason I suspect this is Apple fawns over naming conventions, and "M" was supposed to stand for "mobile" (?). Yes, the studio and mini aren't mobile, but past versions of the iMac and mini used mobile Intel chips. The Studio replaced the iMac Pro.
So maybe they're working on a "P" (for Pro) AS architecture. This is the only reason I can come up with for where the Mac Pro is and is going.
Last edited:
Just maxed out a MacPro here in Aus and its around 1/4 the price of a maxed out intel variant.
Ok there are differences in how you can spec them right now, but I wonder if 2,3 or even 4 of these in a node will be a more efficient solution than one maxed intel for very similar money - then we double triple or quadruple everything.
But still only get to 768GB Ram not 1.5 TB, plus no dedicated GPU/afterburner options either.
Food for thought
Ok there are differences in how you can spec them right now, but I wonder if 2,3 or even 4 of these in a node will be a more efficient solution than one maxed intel for very similar money - then we double triple or quadruple everything.
But still only get to 768GB Ram not 1.5 TB, plus no dedicated GPU/afterburner options either.
Food for thought
Given that the PCI-E Slots don't support GPU's, aside from someone wanting to pay a LOT of money for the possibility of adding PCI-E Storage and/or workflows that absolutely require a PCI-E Card, I really think this new MacPro is going to be even more niche than before.
It's cool, sure, but does not seem overly practical or cost effective unless your workflow absolutely requires a Non-GPU PCI-E card AND you can't put it in a Thunderbolt PCI-E enclosure to pair with a Mac Studio.
I know it was unlikely, but I was hoping the MacPro would support GPU's - even AMD GPU's.
It's cool, sure, but does not seem overly practical or cost effective unless your workflow absolutely requires a Non-GPU PCI-E card AND you can't put it in a Thunderbolt PCI-E enclosure to pair with a Mac Studio.
I know it was unlikely, but I was hoping the MacPro would support GPU's - even AMD GPU's.
It took 2 1/2 years, but they finally did it! This transition really good for my workplace, because schools and institutions and such send us their older Intel Macs to recycle or resell when upgrading to the new M1/M2 Macs! I remember when we got some 2012 Mac Minis that came in boxes for the M1 Mini, obviously coming from somewhere that made such an upgrade.
I'm also reminded of how Apple expected the PowerPC to Intel transition to last two years, finishing some time in 2007, but they were able to complete it at the end of summer 2006 with the introduction of the Mac Pro! History repeating itself in a way (also in a way, the new Mac Pro is kind of like an M2 Ultra Mac Studio with PCI slots and a replaceable/upgradeable SSD).
I'm also reminded of how Apple expected the PowerPC to Intel transition to last two years, finishing some time in 2007, but they were able to complete it at the end of summer 2006 with the introduction of the Mac Pro! History repeating itself in a way (also in a way, the new Mac Pro is kind of like an M2 Ultra Mac Studio with PCI slots and a replaceable/upgradeable SSD).
It’s got an Ultra in it, which is huge. Wafer starts took a huge hit during the pandemic and are still woefully unprepared for the demand N3 has…there are other factors at play besides just how many Apple is looking to sell.
What in the world do you need an afterburner card for?
PCIe isn't just for graphics cards. Presumably the M2 Ultra chip itself has all of your graphics needs covered.
The PCIe slots in the new Mac Pro are there for things like pro audio control interfaces, pro video capture interfaces, additional super high-speed flash storage, etc. Basically you only really *need* a Mac Pro if your career is in professional sound engineering and/or professional video production. As an edge case: professional app development studios may need a newer Mac Pro for continuous delivery / continuous deployment purposes.
But also not worth the development effort with such low volumes. One M2 ultra is at least 16 M2s for MacBook Air or iPad Pro.
C’mon people, they are at the end of the way.
Do an analysis of the latest iOS, wOS, iPadOS, macOS updates: it's half games and customisations for kids and the other half additional finishing of the same things. An example on all Apple Watches, they redone 3 times the gyroscope, but the sensor for glucose is still not there, in fact wOS 10 also runs on S4, which is the same family as S8 and Ultra.
Why don't they implement something serious?
Because they are caught up in the development of Silicon and Vision, and they don't even know what to invent to move forward.
They squeezed the cow as far as they could, and the cow is you.
Do an analysis of the latest iOS, wOS, iPadOS, macOS updates: it's half games and customisations for kids and the other half additional finishing of the same things. An example on all Apple Watches, they redone 3 times the gyroscope, but the sensor for glucose is still not there, in fact wOS 10 also runs on S4, which is the same family as S8 and Ultra.
Why don't they implement something serious?
Because they are caught up in the development of Silicon and Vision, and they don't even know what to invent to move forward.
They squeezed the cow as far as they could, and the cow is you.
I would say that they never expected to modify their System-on-a-Chip to be expandable. Each step is taking a lot of extra work to make everything work together.
Its not I want to add afterburner - its what was able to be added under Intel vs what can be added now - sorry for confusion
Why wouldnt you think so? As long as the device is detectable it should be usable, it’s the same way how once Asahi has thunderbolt working fully you should be able to use eGPUs on M* machines under linux. The barrier is the TB controller. Support is a work in progress right now. In this case it’s straight off the exposed PCIe lanes without the TB layer, nearly the minute Asahi is working on a MP you should be able to use dGPUs, there are already drivers for ARM under Linux for the cards
For some AI researchers, this will be a total bargain. A "no-brainer, buy it now deal".
For something like $10K you get 192 GB of RAM connected to a pretty decent GPU. Look at the competition. Where else can you get a GPU with 192 GB of VRAM for $10K
Yes, with Apple's unified RAM all the RAM is accessible to the GPU. This Mac Pro can actually train a GPT3-like large language model locally on the Mac.
You can not buy, even, a low-end Nvidia A100 GPU for $10K and even if you could you'd still need to buy a computer to drop the A100 into. Apple sells you the entire machine for half the price of a single A100.
Will I buy it? No. I can't possibly afford it. I will rent time on Google's cloud. Renting is cheaper than buying if you only need a few hours a week.
OK, you might argue that a Mac Studio is better for this use case. I think it comes down to cooling. Can a Mac Studio run 24x7 at 100% full power and not over heat? A MP will have better cooling than the MS. But how much better? We don't know.
TensorFLow and Pytorch both now run on Apple Silicon. I own a "lowly" M2 Pro and it seems to run machine learning tasks about as fast as a low to mid range Nvidia GPU, but the M2 can have more VRAM and is cheaper then a high-end Nvidia GPU. Apple is competitive with Nvidia at the $2,000 price point. Apple just might win at the $10K price point. But Apple has nothing to office at the very high.
For something like $10K you get 192 GB of RAM connected to a pretty decent GPU. Look at the competition. Where else can you get a GPU with 192 GB of VRAM for $10K
Yes, with Apple's unified RAM all the RAM is accessible to the GPU. This Mac Pro can actually train a GPT3-like large language model locally on the Mac.
You can not buy, even, a low-end Nvidia A100 GPU for $10K and even if you could you'd still need to buy a computer to drop the A100 into. Apple sells you the entire machine for half the price of a single A100.
Will I buy it? No. I can't possibly afford it. I will rent time on Google's cloud. Renting is cheaper than buying if you only need a few hours a week.
OK, you might argue that a Mac Studio is better for this use case. I think it comes down to cooling. Can a Mac Studio run 24x7 at 100% full power and not over heat? A MP will have better cooling than the MS. But how much better? We don't know.
TensorFLow and Pytorch both now run on Apple Silicon. I own a "lowly" M2 Pro and it seems to run machine learning tasks about as fast as a low to mid range Nvidia GPU, but the M2 can have more VRAM and is cheaper then a high-end Nvidia GPU. Apple is competitive with Nvidia at the $2,000 price point. Apple just might win at the $10K price point. But Apple has nothing to office at the very high.
Last edited: