Apple seems to have painted themselves into a corner by being perfectly reasonable. You see, they recently updated the Mac Pro tower, and it turned out to be the Mac Studio Ultra in a bigger box. The pci-e slots inside are barely of any use since GPUs are not supported, and the machine is inherently un-expandable; you can’t add more RAM, for instance. So, while this machine is extremely powerful, it’s by no means an open-ended system you can expand to fit your extreme workflow – for instance, trying to do fluid dynamics simulations or something. So, basically, the biggest Mac is great at doing normal Mac things, such as movies and audio, but it doesn’t extend into high-end scientific or engineering workflows. Is that a problem? I don’t think so, and let me explain why.
What Apple did right was design a range of machines that covers their professional and prosumer user base; people who make videos in Final Cut, make music in Logic Pro, or edit photos in Lightroom and Photoshop. They made excellent laptop range that covers everything from writing articles and checking Twitter in Starbucks, to the most complex software design, or audio, video and photography work. Then they made a desktop range that covers all of that and extends deeply into professional studio use, and I don’t think they left even a fraction of their traditional or potential user base uncovered. Heck, even I got a desktop Mac, which I resisted so far because all they had either came built into a display, overheated under load or cost extreme money while offering no benefits over what I had. The drawback of the current Mac lineup is that it doesn’t extend endlessly, and some, like Linus, will whine over that, while I might offer a more constructive approach.
You see, there’s only as much you can expect a desktop machine to do, and M2 Ultra actually exceeds this expectation by far. In order to exceed this ceiling, Apple would have to completely redesign the system that’s perfectly good for 99.9999% of their user base, in order to cater to the affectations of those who actually need a supercomputing cluster but aren’t technologically savvy enough to realize it. In order for Apple to try to artificially meet this almost non-existing demand, they would have to create a completely new architecture centered around a passive bus motherboard and Max/Ultra daughterboard cards that connect over this bus the way two Max chips are edge-connected to form the Ultra; in essence, combine the infinity fabric and PCIe technologies and make it work. The engineering overhead would be immense, and they would only sell a handful of those machines because the demand at such a high-end of computational needs is very slim.
Alternatively, those who actually want to perform complex computations should make their software cluster-friendly (basically, dump workload packages onto a NAS stack, and have the cluster nodes run workload processors that pop packages from the workload stack, process them and push them onto a result stack, and when that is done, have some script go through the results, check their integrity and assemble it all into the end product). Then you could have multiple Mac Mini or Studio devices connected to the LAN, processing the work you give them, and you could extend this infinitely, and the demands on an individual node would be such that you could optimize it for cost-effectiveness by buying the base Mac Mini or Mini Pro devices by the truckload. This kind of work is usually done on Linux nodes, but a modern Mac is so power-efficient that it has genuine advantages for cluster applications.
So, basically, the Mac lineup is truly powerful enough, and it allows for open-ended design for those who need endless computational power; it’s just not open-ended inside a single chassis, which is only a problem if you have unrealistic expectations, or a workflow that is poorly designed, by not breaking down the job into manageable components. Honestly, I could write such a cluster setup myself in less time than it would take me to start whining about lack of power and tasks taking too long.
Indeed, Mac Mini clusters are a thing for some time now and they were upgraded to Apple Silicon soon after M1 release.
And if it can be done on Enterprise level, it can surely be done in a lab level as well.
The only problem I see is if you want something like this, but don’t actually need it and then whine that’s it’s “hard, complicated and expensive” – something like Linus.
But if you do need it, you most certainly won’t whine that buying 100 Mac Minis “is expensive” and that setting up cluster is “hard and complicated”.
Regarding Mac … only problem could arise from the fact that Mac Pro was redesigned from ground up to be easily upgradable and they bragged a bunch about it – and now suddenly it is not any more.
Even if you max out current Mac Pro – you might want to benefit from new gen CPU or GPU in some high-workload environment few years down the road – which now you can not – you have to buy new Mac Pro which certainly raises ownership cost – good for Apple, not so good for studios.
That will be much more psychological than actual issue, since those customers have money to buy new Mac Pros if they actually need it and new Mac Pros are so much faster it is ridiculous.
However, it might have been better if Apple didn’t transform redesigned, upgradable Mac Pro to a thing upgradable as iPhone, instead, they could have simply introduced Mac Studio Pro and slowly phase out Mac Pro.
Not that I care, I use MacBooks for some time now, since it is convenient to pick it up and use it anywhere and even base M1 Pro model provides ridiculous amount of power for my needs and only upgrade I might consider is external NVMe drive.
I think the Mac Pro was released to show it’s obsolete; basically, it’s cancelling it by showing and not saying. Sure, it’s nice to have a machine where you can buy the base model and then pop in more RAM and SSD, and that’s definitely something I would qualify as a drawback of the current design. Also, the inability to pop in a powerful Quadro card if you need it for professional work is a minus, and because of that some workflows will simply not be done on a Mac. In my case, the only things that don’t run on the Mac are the games (specifically, Witcher 3) and that idiotic Croatian government smartcard digital signature app that was written by someone’s cousin and works only on Windows, despite being formally supported on Mac and Linux. It does run, but doesn’t actually open the pop-up that lets you select the certificate in the web browser, and therefore doesn’t work. It’s not a Mac or Linux problem, it’s a government corruption problem.