For those hoping for an M1 based server, this is the definitive signal that will not happen.
…that it will not happen via an add-on “server” program.
But TBH it’s hard to see how servers could be much of a market for Apple. People use the cloud more than local (“departmental”) servers these days and the cloud back end belongs to Linux, typically with a provider on top (AWS etc). Apple’s strengths lie elsewhere.
It’s done to death that their web services are of dubious quality, and they have got better. However you don’t have to go very deep or use services for very long to find some very sharp edges.
You can have a 5U server with lower power usage that is a LOT cheaper than 1U power hog.
Obviously at AWS scale this is abstracted away, but I think you are still paying a lot for the power you consume.
So not being able to specifically use Apple silicon isn't a big deal.
However it maxes out at 16gb ram (a Nuc 9 goes to 64) and virtualisation isn’t quite there yet I don’t think. Fusion says it supports the M1 but I don’t get good results.
I’d like esxi or Proxmox on it, but could learn to live with it as is I suppose.
Running a GUI when headless seems a bit crazy.
There's many hundreds of thousands of windows servers that run a GUI.
Not saying I like it; but, it's a thing and nobody seems to care that much.
This is what I use for my file server:
https://ameridroid.com/products/odroid-hc4
Runs Debian (Armbian). Provides file shares over SMB, Syncthing, S3 (Minio). Also does some monitoring of other systems via Prometheus scraping of node-exporter. Everything except Samba runs in containers.
With two mirrored 6TB HDDs it has no problem keeping up with the household 1GbE network. Quad A55 cores, 4GB with Zram and it has all the power I need.
It can't live-transcode video above about 720p without stuttering, but then, most of those ARM boards can't either, so I just have to make sure whatever client I'm using can handle native formats for all my media.
Not sure how it is on power use, but I much prefer managing all that stuff on one machine.
They usually get dropped or ran over or screen hinges broken or something like that. Long before the actual hardware inside gets a chance to fizzle out.
They have a lot of advantages in that they are low power, have built-in UPS, small, have a keyboard and display. The disadvantage is that you don't get a lot of storage options. I have some USB 3 attached storage, but I wanted something a little different this time.
Most recently my RPI 4 8GB serving as media server got replaced by my girlfriend's Asus laptop. She knocked it off a table and the corner of the display got smashed in. Now it is running Linux and Kodi, among other things. Can barely see it sitting there under the TV.
A bit better at 4K out then RPI 4, but not by a whole lot. Don't do any transcoding on it, though.
My "Big Servers" for my home lab are 3 ancient HP 1U rack servers. Built a small horizontal rack so they "hang down" instead of sideways and they take up almost no room. They run some high efficiency Xeon processors so I can get away with running them off of one household outlet. I only leave one on and spin up the other two when I need the extra capacity for some project.
Dual socket 8 cores and 72GB of RAM each.
Figured out that if you go and look at Vmware hardware support and get a server that is _just_ one generation older then they are willing to support than these things are dirt cheap on the used market.
If they can't run the latest version of Vmware then nobody wants them. Linux don't care, of course.
I think I spent a total of 600 dollars for all 3 servers, but it could of been a lot less. It's been a while.
I hate dealing with server-grade hardware because the lights out stuff is insecure (100% get one with a dedicated LO network port), dealing with firmware configuration is tedious, take forever to boot up, and dealing with the raid controllers is misery. But it's the cheapest way to get a ton of capacity. Especially if you have a project that wants to use IPMI and such things.
Licensing requirements for VMs and lack of server hardware that can be put in datacenters are both obstacles to target Apple Silicon. Not insurmountable, you can build enclosures for a rack of Mac Studios & Mac Pros and orchestrate VMs with Hypervisor framework, but it's a large enough obstacle that major cloud providers don't offer it and that pushes large costs onto the CI providers, which then get pushed to the folks trying to build tooling that works on M1 Macs. That includes the open source ecosystem.
The alternative is cross compilation, either cross-arch or cross-platform or both. All of these options are brittle and complex.
Speaking from experience here, I've had to follow issue trackers for: programming language teams shipping parity for M1, major libraries (think of foundational libraries like GRPC's libraries for each language) not yet shipping binaries for M1, and developer tools and OSS teams producing binaries. All of this needs to come together, and it hasn't yet.
The whole ecosystem is straining under the weight of this complexity to target engineers' laptops.
https://aws.amazon.com/ec2/instance-types/mac/
You can choose from either x86 or M1 machines.
It's been nearly two years since the Apple Silicon kits first shipped, and the only instances you can purchase from AWS are metal instances, at approx. $10k/year/instance.
That's not accessible to open source/free software projects and not affordable or efficient to utilize for small projects and highly intermittent workloads like CI.
I wish MacOS licensing were available for Graviton and ARMv8 generic, that would be incredible. It isn't, and the costs to support Apple's platform is a substantial barrier for many projects.
Hobbyists? I spend $700/year to have access to a maker space. I spend that much on woodworking tools annually (if not more). Hobbyist doesn't mean free.
That said, Apple isn't going to sell chips these to anyone else, period. And everyone right now is hustling to match the same levels of efficiency in newer designs, and they'll get there sooner or later. So "Apple Servers" are a completely moot point, but power efficiency? Yes, it literally matters more than ever, across every spectrum of the industry, and it's why the M1 was something people cared about anyway.