It's really not. A nVIDIA graphics card will do it better, for MUCH less, and unlike Apple's M-series chips, nearly every AI model is built to run on nVIDIA.
Can you run some models locally on this? Sure. Is it ideal? Not by a long shot.
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.
Mac Studio With Massive 512GB RAM Facing Two-Week Delivery Delay
- Thread starter MacRumors
- Start date
- Sort by reaction score
Can you run some models locally on this? Sure. Is it ideal? Not by a long shot.
The problem with those gpus is limited dedicated ram
I can't imagine they were expecting to make many of them. The base model will be the ones they have stock of, I would speculate that's part of where the price comes from, because they have to be made to order to a certain extent.
All they need is to sell 1,000 units of these to make $15 million. That's pretty crazy. For a higher spec M4 Max MacBook Pro ($4k), they would need the equivalent of 3,750 units. The numbers must be crazy for Apple. I forget about them sometimes
I don’t have any problems with the Mac Studio. It’s not even a first world problem. You can either afford it or not. On the delay of one week that’s not what I am talking about as far as depreciation goes - seems obvious I am talking as each new model provides more the prior models drop substantially.
I wasn’t referring to the 512GB RAM version. The Mac Studio starts at $2,000.
Only for what fits in memory.
For most people that is limited to 24GB for those who are lucky enough to score a 5090. And price of that card along with the supporting PC will end up cost as much or more than the Mac Studio.
Sure, it will be much faster for your little 7B models, but you'll not be using the 70B models.
Furthermore, if you want a new A6000 Blackwell-whatever-Nvidia will call it, that alone will cost more than a Mac Studio.
And all those systems will be noisy, require air conditioning unless your in the Yukon, etc.
In reality, the Mac Studio is just better for someone who wants something to fit on their desks.
If you need more than what the Mac Studio, then it's just cheaper to rent processors remotely.
And it's what the retailers like MicroCenter and B&H will stock, and occasionally Costco when the wind is blowing in the right direction.
Speaking of which, Microcenter has the binned M4 Max Mac Studio for $1700:
Apple Mac Studio MU963LL/A (Early 2025) Desktop Computer; Apple M4 Max 14-Core CPU; 36GB Unified Memory; 512GB Solid State - Micro Center
Get it now! Powered by the most advanced Apple silicon, Mac Studio is the ultimate pro desktop. With massive CPU core updates, mind-bending graphics, increased memory, blazing-fast Thunderbolt 5 connectivity, and next-level AI capabilities, Mac Studio is the beast thats built to deliver.
And get this, they have the base-RAM binned M3 Ultra for $3400:
Apple Mac Studio MU973LL/A (Early 2025) Desktop Computer; Apple M3 Ultra 28-Core CPU; 96GB Unified Memory; 1TB Solid State - Micro Center
Get it now! Powered by the most advanced Apple silicon, Mac Studio is the ultimate pro desktop. With massive CPU core updates, mind-bending graphics, increased memory, blazing-fast Thunderbolt 5 connectivity, and next-level AI capabilities, Mac Studio is the beast thats built to deliver.
That latter one beats the EDU price by $200!
I suspect that max config has been selling at a very high rate.
Most people have no interest or use for it, but for those who are dabbling with locally hosted AI and need the large amount of GPU memory to run them - there is no real alternative south 6 figures. Nvidia 80 GB GPUs are something like 10-20k USD each, and you'd need 6 of them along with multiple systems to host them in to build something that can run what the new Studio can.
You could maybe work with a cluster of machines with multiple cheaper consumer 3090s or 4090s or 5090s but good luck finding them; you'd need 20 of them at ~1-2k USD each (for 5 year old RTX3090s) and a bunch of PCs and/or crypto mining PCIe multiplexors to use them.
Last edited:
Considering how flat footed Apple was caught on the application of AI, maybe the only silver lining on that front at the moment is that they lucked into having gone all-in on an archictecture like AS. If they aren't working day and night to fill up datacenters with these Ultra SoC's then I don't know what to say.
You say they've been caught flat footed, i say they've been doing a lot more with ML/AI than others have for a long time, they just didn't hype it as much. They've been stuffing M series servers into their datacenters for a few years now.
Half a decade ago they were doing on device AI photo classification, etc.
Apple's AI features are mostly neat stuff it does in the background rather than chat bots.
Unified memory being the future has been obvious to most of the big players in tech for a decade plus by now (because of the numerous benefits), its just the tech has taken time to catch up. Apple has been uniquely positioned to take advantage of this first as they aren't building GPUs that can outrun the shared memory architecture at the moment, and for some workloads (ai, and others that apple focus on) memory size and access to both CPU and GPU without copying is more important than speed.
Even 15 years ago or more there was industry talk of moving the processing to the memory chips (may still happen), so far they've just moved the memory much closer to the CPU/GPU.
I agree with all of that. "Flat footed" is hyperbole, but I think it's fair to say that Apple is behind the curve compared to the other big AI players. You're totally right that the were pushing the tech forward for years but they still somehow got caught off guard by the sudden application of the tech by the likes of OpenAI. If it turns out that AI as it is marketed today turns out to be an over-hyped nothingburger then Apple's steady hand will have served it well. But in reality, while I expect a slight hype correction, the generative AI push will be more boom than bust.
Obviously I'm not allowed in the rooms where these things are discussed. But I find it hard to ignore Apple's current predicament and draw some conclusions.
Knowing how on-the-ball Apple has been with AI, I bet they had exactly ONE ready to ship on day one. "For AI? We have image playground. What else AI would anyone want?"
And from all indications, it seemed Apple was not even interested in delivering such a beast. This while people were cobbling M4 Minis together to run LLMs as Apple wouldn't even deliver an M4 Max option of the new Studio unit a year later.
Apple better not sit too comfy, as Nvidia is said to have its ARM-based solution on the market later this year. With Cuda and all the trimmings.
And from all indications, it seemed Apple was not even interested in delivering such a beast. This while people were cobbling M4 Minis together to run LLMs as Apple wouldn't even deliver an M4 Max option of the new Studio unit a year later.
Apple better not sit too comfy, as Nvidia is said to have its ARM-based solution on the market later this year. With Cuda and all the trimmings.
Last edited:
That’s true but the Nvidia solution will probably require Windows unless one is building a Linux machine from scratch.
I ordered the 512 GB w 8 TB SSD 2 weeks ago and it said "Delivers March 24 - 27th", so I was expecting it any day now. Sadly I log on today to see if's preparing to ship and it's been updated (with no email or notice) to say "Delivers April 8 - 10". So now 2 more week delay. 
.
.