> They could reuse the high-end Mac Studio or similar chips and only have a custom board and chassis suitable for server use, with additional IO and redundancy.
Also what would be the benefit of running Xcode cloud on Apple Silicon? Today cross compilation is the norm so the architecture doesn't matter. They would have to make a new datacenter version of their chips without the GPU and accelerators that aren't used for CPU heavy tasks.
> they literally make the Mac Pro in rackmount form for render farms and so forth.
The Mac Pro has a rack mount version is not for the server room. People use racks in audio and video production too and that's where you will probably find most of the rack mounted Mac Pros. You can probably fit 2 CPUs and 4 - 8 GPUs in the space a Mac Pro takes in your datacenter. So why would you choose a Mac Pro over that?
I have four Mac minis racked up for this, with two use cases: 1) iOS CI/CD and 2) some computer vision stuff using Apple’s Vision framework.
No complaints, but I obviously wouldn’t run anything that didn’t need to be on a Mac on these systems.
BTW, you can rack mount twenty Mac minis in the footprint of a single rack mounted Mac Pro (there’s a 1U mount that takes two and they’re small enough to mount on both the front and the back of the rack). So 20 M2 Pros per 5U.
They’re of course not a sysadmin’s dream but they do tend to stay up and Ansible works fine.
It might. Mac sales have been declining for several years now.
I’m not so sure about that. I think the Apple Silicon thing added a big boost.
But I’m often wrong. I’m sure there’s numbers, somewhere.
https://www.macworld.com/article/2130070/mac-sales-slump-rev...
But I never found Mac servers particularly compelling. We used to have a couple of XServes (and the overpriced RAID disks), for years in our rack, where I worked. We never bothered using distributed builds, which was probably the only unique service they offered.
In Apple's case you also have the people who upgraded early for the M1 performance boost, and they aren't likely to toss out a new computer unless they are doing cutting edge tasks.
Anecdote wise, my corp forced a refresh in Q4 so I expect a crest is coming in 2024.
- Highly energy efficient non-mobile computers (mini, pro, etc).
- They already run a huuuuuuuge server park on AWS for themselves
- They've been (successfully) transitioning their revenue towards "services" for almost a decade
- I suspect they want a piece of the LLM market as well.
- They acquired Fleetsmith for device management in the enterprise
- They implemented nested virtualization in the M2+
- The cloud is highly profitable, and I suspect they want a piece of that cake
- They spent a lot of resources creating a boot policy that allows booting other/unsigned operating systems on Apple Silicon. (asahi). This is not present on iOS
- They already offer "cloud runners" for XCode (cloud). This is a very contained service, to start offering cloud services
- The timeline for iCloud platform:
- 2016: Azure -> Google. 2018/2019: Google -> AWS. This $1.5B contract will end this year. It seems like a prefect moment to switch from AWS to their own platform and later-on offer more generic services to companies.
- They need to 4T or 5T marketcap somehow. Options are - LLMs, Cloud servers, the Enterprise, (Vision), or something completely new, but the Vision Pro is the last major category under Tim Cook, so I don't think anything new will happen for consumers.
All in all, it seems like Apple will go back in the (cloud) server business.
As far as the complex list of reasons:
- Correct.
- Do they run the server park, or AWS?
- "Services" is a word Apple uses as shorthand for Wall Street types to mean "monthly subscriptions to consumers", it doesn't mean "things we charge for", because at that point it has 0 meaning
- Why would anyone run an LLM on an M-series server?
- Why would Fleetsmith help manage servers? It's MDM. You don't need or want provisioning profiles on servers.
- Nested virtualization makes sense.
- Cloud isn't highly profitable, it's highly revenue-able. Profits are declining as the titans bash it out.
- They didn't spend a "lot of resources"
- Xcode Cloud is in the cloud - but they're already offering it.
- The timeline for iCloud elides a whole lot of time.
- They don't need a 4T or 5T marketcap and those aren't the only options.
You know too much about Apple and it's blinding you because you can connect all the info you know about it to any other arbitrary trend. Been there, that's why I'm comfortable being overly concise about it.
Services is anything software as far as I know.
If they run their own servers, they'll have a large number of gpu core which could be used for LLMs. From what I understood is that M* is pretty good at them. That way they're not helping NVIDIA grow more as a competitor.
My point with Fleetsmith is that they're looking at corporate use.
Maybe not a lot of resources for them (boot loader), but they deliberately allow booting other operating systems.
I firmly disagree about cloud profitability. I read numbers between 30% and 60%, although I also found 2.6% for GCP and -8% for Azure. Not sure why they're so far apart. Amortization shouldn't be the issue. Perhaps "R&D"
Last part, you might be right ;-)
Cloud customers want good prices and well-supported hardware. On the server, Mac is neither of those really. Apple's hardware pricing would need a complete revision to stay competitive, and their software policy would need to open up even more to enable better API coverage, proper virtualization and compelling hardware compatibility. Even if Apple did that, it would merely be enough to keep MacOS a second-class citizen in the world of server OSes. The existence of OpenBSD and Linux almost obsoletes any real-world application of a licensed server OS.
Apple can go back into the server business if they want, but the server business isn't begging for more overhead or grudge matches between vendors. Apple cannot turn the server industry into a trillion-dollar opportunity without reversing policy entirely, on both the hardware and software side of things.
ARM builds can be done on x86, but what about the automated tests after? Those have to be run on ARM when you target AS and everyone including Apple themselves needs something to do that.
The rackmounted Mac Pro is actually intended to be used amongst other rackmounted audio equipment in Studios. FWIW.
That it can be used for both is a nice accident, but it's not well optimised for a render farm. Not that this matters really as it can be used this way so they can still get access to this market.
There were at least a few threads on Apple’s forums where people were redirected to community-supported software.
There is a fairly strong use-case now with robust software, but I feel like Apple wants to maintain an illusion that their developers are small mom-and-pop shops with a Mac Mini in corner of their farmhouse.
But I doubt they would do this well. It would be way, way too expensive and probably not have upgradeable RAM.
I wonder what kind of software/OS they run on those machines.
The RAM being part of the chip seems like a limiter to scaling up to a server model with a lot of cores and RAM in a single blade.
Not saying this is not all doable just that it probably requires design and investment they otherwise do not have any desire to do.
1. less waste. The entire case is being thrown away immediately by these volume consumers. All the packaging and what not is also wasted.
2. easier/cheaper deployment. Companies are already paying their employees to design replacement cases, control boards, etc. and getting them built at relatively small scales. Then they're paying them to shuck and transplant from the Mac Minis. Apple could probably charge a >50% premium for their solution vs the equivalent number of Mac Minis and companies would jump on it.
Additionally, at least in my experience, the headaches and fixed costs of deploying these machines means companies refresh them far less often than other hardware. It's totally possible that companies would refresh more frequently if Apple just sold something easier to deploy.Nearly all of that backend Apple infra runs on Linux and (formerly?) Solaris.
As for why Oracle: I think they said something along the lines of Oracle indemnifying them -- or being best equipped to mount a defense -- in a software licensing lawsuit.
Unlike companies like Google, Meta and Microsoft they are institutionally afraid of over-hiring and losing their culture. This is a conservative tradeoff where they may lose the ability to grow seemingly "small" markets (small=anything smaller than the iPhone) in favor of not losing their culture (becoming a Google or Meta which just squanders every opportunity)
Apple themselves would make good use of the Xserve for their own internal use rather than buying Dell/HP servers like they currently do. Also things like Xcode cloud are currently running on x86 commodity servers, in the future Apple likely will stop supporting x86 altogether and it makes sense to run it on AS. They could reuse the high-end Mac Studio or similar chips and only have a custom board and chassis suitable for server use, with additional IO and redundancy.
And for those saying that Apple isn't interested in large deloyments, they literally make the Mac Pro in rackmount form for render farms and so forth. The Mac Mini has 10G ethernet and the formfactor purposely hasn't changed. The chips are efficient enought that you don't need screaming loud 1U fans and they could definitely create a quiet 1U Studio suitable for both dense datacenter deployments and a TV studio technician's rack.