Just when we thought that the M3 was in the ground and that Apple had quickly moved each Mac model to the fresh and new M4 processor, in Rolls the high -end Mac studio, equipped with the M3 Ultra M3 processor impossible to predict and without prediction.
It is true – in March 2025, 16 full months after the arrival of the M3 Pro and M3 Max chips on the stage, Apple introduced a new chip into the old almost retired family. And this means that the new Mac studio is available simultaneously in M3 and M4 varieties – with the M3 model being the high -end configuration
So confusing. I admit that I was a little disconcerted when I also heard the news. But the more I think about it, the more I grabbed the fact that the Apple chip strategy is not as simple as it wishes. And if I told you that the M3 Ultra is not quite an M3 Or A M4 chip? Follow me, friend, and see how deep this rabbit burrow is.
M3 fence sale
Here is the story as we know it: Apple and its flea manufacturer, TSMC, were rightly proud when the M3 and A17 Pro chips arrived in 2023 as the first chips made by a new 3NM peak manufacturing process. But it turns out that TSMC was already working on a more efficient and economical second generation 3NM process, which makes this first generation a little architectural dead end. Therefore, everyone expected Apple and TSMC to quickly turn from M3 to M4, which is based on the second generation process and more efficient all around. This is why Apple has spent the last months updating almost all Macs from M3 to M4.
Petter Ahrnstedt
Except that … there is always the M3 iPad Air and the iPad Mini A17 Pro, both of which use the cutting rate versions of this old process. No one outside of Apple and TSMC seems to know if there are still parts of the old process that runs somewhere or if there is a large BAC of M3 and A17 parts that Apple will know to meet any requests for devices with the old chip architecture.
It was quite weird, but the announcement of the M3 Ultra is something else. It's a new Puce, not a relic of the past. This is a whole generation out of step with Apple's current strategy, has thawed and has deposited an audience without distrust like a kind of Silicon Austin Powers.
Or at least that's what you might think if everything you see is the big number 3 next to the capital M., but when you look more closely at the M3 Ultra, you start to realize that everything is not as it seems.
M3 or M3.5?
The Ultra and M2 Ultra M1 chips were mainly only two M1 M1 MAX and M2 Max processors, respectively, glued with Apple Ultrafusion technology. Their specifications were simply duplicates of smaller tokens: twice the CPU nuclei, twice the GPU nuclei, twice the neural engines and twice the maximum RAM.
But this is not the case with the M3 Ultra, this is probably why it took a lot more time in Apple to release him than his brothers' chips. (The M2 Ultra followed the M2 Pro and Max after only a few months, so 16 months, it's quite an expectation.)
The M3 Ultra chip is very modeled on the M3 Max and offers twice as much CPU and GPU nuclei, neural engine nuclei, etc. But look more closely: the M3 Ultra offers a maximum of 512 GB of RAM, four times The maximum 128 GB of the M3 Max. And the M3 Ultra supports the faster specification of Thunderbolt 5, not the oldest Thunderbolt 4 of the rest of the family.
This means that the M3 Ultra is an intermediate chip, mainly based on the old M3 generation, but with a few high -end additions that push it to the M4 in terms of capacity. In my conversations with Apple representatives, I had the impression that these modifications, including the upgrade to Thunderbolt 5, had an impact on the release date of the chip.
So why call him an M3 and not an M4? I think that Apple has fairly strict internal rules that determine the name of fleas, and they are probably starting with the design of processor hearts. A glance at the monochers performance of the M3 Ultra and you will see that it works more or less like any other M3 chip. Of course, you would Never Buy an ultra M3 and use a single processor kernel – the whole point is that you have 20 or 24, and 60 or 80 GPU cores.
Foundry
Speaking of maximizing things, the M3 Ultra is designed to do so. Because it is mainly the architecture of previous generation chip, each part is slower, but if you load all these CPU and GPU cores and enjoy a huge pile of RAM and the high bandwidth of Thunderbolt 5, the M3 Ultra Mac Studio will perform rings around a M4 Max Mac Studio. It is a faster chip – when used to the maximum.
This is why the M3 Ultra Mac Studio is really a product designed for a relatively small audience. Even most professional users will find it difficult to make the M4 Max chip feels slowly, but for certain applications – video, rendered, science and AI – the enormity of the specification sheet of the M3 Ultra is worth its very high price.
As Ben Thompson pointed out this week, an Ultra Mac M3 studio with 512 GB of RAM could execute the quantified four -bit version of the Ultramodern R1 Deepseek R1 model on the device. Even at a price of tens of thousands of dollars, the most high -end Ultra Mac M3 studio could be a good deal compared to expensive NVIDIa networks – if you need to manage a huge model of AI locally. Almost no one wants that … But if you are an AI researcher, you could certainly.
Some open questions
However, the arrival of the M3 Ultra is bizarre enough to give the impression that we all miss part of the situation as a whole. If there is no other shoe to drop, why has the Mac Pro based on M2 Ultra not obtaining its own ultra chip M3 bump?
Of course, this could mean that the Mac Pro is just a Lagcard, and it will appear in June with the same chip inside. Or it could mean that Apple does not care about the Mac Pro and that it aspires to the fjords.
But consider this: Apple suggested that there will not necessarily be an ultra chip in each generation of chips in the M. OK series, but … and if there is something else? A few years ago, there were rumors of a still high -end chip for the Mac Pro, but it would have been canceled. I wonder if perhaps the future of Mac Pro is hidden at sight, and the M3 Ultra is because Apple will use its high-end flea design bandwidth in the M4 or M5 generation for something a little bigger, designed just for the Mac Pro.
(A huge chip should almost certainly be inside the Mac Pro, because I doubt that the Mac Studio speaker and its cooling system would be able to manage such a large and hot chip.)
I do not know if a sophisticated ultra-high chip for the Mac Pro is in the cards, in this generation or any other. But if the M3 Ultra Mac Studio already has intriguing implications to perform huge models of AI locally, imagine what could be activated by an even greater chip with even more on -board memory. It would be a set of features worthy of the name Mac Pro.
