$599 seems like a lot to me. You can get numerous older, much more powerful Mini PCs (e.g older ThinkCentre Tiny series) or even a base brand new M4 Mac Mini for that kind of money.
Admittedly, the 10G interfaces and fast RAM make up for some of it, but at least for a normal homelab setup, I can't think of an application needing RAM faster than even DDR3, especially at this power level.
A base Mac Mini (256GB/16GB) would cost me €720 while a Minisforum MS-R1 (1TB/32GB) would cost me €559 (minus a 25 euro discount for signing up to their newsletter if you accept that practice).
Price to performance the Apple solution may be better, but the prices aren't similar at all.
Upgrading the Mac to also feature 1TB of storage and 32GB of RAM, the price rises by a whopping €1000 to €1719.
Yes, I was basing my estimate more on the US price of $600, which is basically the same, if you buy both at MSRP. (However, that means the Mac does not have 10G)
I did not realize the EU versions were that much more expensive.
I do agree about the RAM/storage prices though. It's only worth it if you want the raw power, where the Mac handily beats this.
Mac M4 have also the problem that you can't install whatever distro you want too, even if it is cheap. When they'll get out of support, there's no upgrade available anymore
The MacMini will still be running in 5-7 years, reliably, and still have updates from Apple on the macOS side. It will still be running in 10+ years too should you keep it that long.
MinisForum makes disposable hardware. We used to use them for TV computers at work, and while they are cheap, they are fidgity with hardware and drivers, come with hacked-Windows Enterprise installed by default, and generally last for about 2 years before they hit the recycle pile.
559 vs 720? That's literally like a few coffees. I went to Amsterdam (assuming you're dutch) and I paid 5 euro for a coffee.
Go for the Mac Mini, the hardware incl thermal is also built exceptionally well. That's why you still have 20 year old Mac Minis still running as home servers etc.
If you're spending 170 euros on coffee then you're either abnormally rich or abnormally bad with money for a Dutchman.
Without the ability to upgrade either storage or RAM, a 256GB SSD with 16GB RAM is quite useless for a home server. Minisforum doesn't offer any options with that little RAM and storage it seems (you can pick between barebones and 1TB models).
The bare minimum spec for the Mac Mini sits at an interesting price point, but if you use it for any more than the bare minimum it'll be pretty restrictive with how memory-hungry macOS has become. No Linux support to speak of also makes for a rather mediocre home server experience.
One interesting part I found out of Apple's European pricing is that after currency conversion and subtracting VAT, the European price is still equivalent to $700, which is $100 more than they charge within the US. Looks like a 1/6th price increase is all you need for consumer rights!
I spend about ZAR 1200 (or 60 EUR per month) on coffee at home but who knows with all my cappuccinos. It's not really cheaper here in South Africa. But thanks, you made me look at my own coffee consumption now and it's always good to know!
Indeed macOS is a bit memory hungry but... unified memory, the sheer speed those chips can move data around is ridiculous. And macOS is a proper workstation Unix.
You're right - it's not ideal for headless. But there are ways. Still less painful than running Windows as as server.
> a 256GB SSD with 16GB RAM is quite useless for a home server
Not that I would buy it new, no, but 16GB for a home server can be quite fine. If I didn't have the 2x3 TB NAS disks (sounds ridiculously small now, right) - that would actually be enough.
I feel like the "that's just a few coffees" metric is getting out of hand. By this metric, my current work laptop, purchased used from a local used reseller, was "a few coffees".
Also, I'm surprised how often on here I see people argue about price differences that are literally as I spend on entire computers.
> That's why you still have 20 year old Mac Minis still running as home servers etc.
I often see statements like this made as if it's an exceptional characteristic of Macs. I've found that almost all computer hardware I buy has made it 20 years, though. Sure, a hard drive or something dies every once in a while, but most stuff gets retired because I just don't care to use it anymore, not because it doesn't work.
Exactly. Of all my hardware since 2003, which includes 5+ different GPUs that were mining and later training AI models almost non-stop the only things that stopped working and not just discarded for being too old/slow are 2 OCZ 2 SSDs which my guess would be had a bug in their firmware that caused a lockup.
In some cases yes, but having used Macs for decades and also working in companies with Macs for all the developers one thing is clear; these things don't easily break. Built extremely well.
(One exception being the GPU issues a few years back though on Intel MBP's)
Other than that whole butterfly keyboard on their laptops for several years, culimnating in a $50 million class-action lawsuit for laptops between 2015 and 2019, totally!
No, wait, there was also bend gate and the iphone battery thing as well.
Uhhhh something about airpod ANC crackling.
There's more I'm forgetting.
God why do I keep buying Apple shit. I'm in a toxic relationship with an abuser.
Agreed. I can definitely see the Minisforum being far more cost efficient if you're mostly doing high speed networking transfers, while the Mac is more cost efficient if you need more raw power.
I just picked up a NAS - a ugreen dxp2800 - for £300. It has 2x nvme slots and 2x 3.5 bays. It’s x86 so if you don’t like the ugreen os you can change it.
It runs docker (supports docker compose) and vms and has the usual raid stuff.
They also do an arm version for half the price but I wanted the intel gpu for transcoding.
Not sure about the CPU performance being much more powerful for some shit-stained NUCs found on ebay, but one selling point for these minisforum machines are hassle-free dual 10G interfaces which are required for decent cluster performance - see ceph or proxmox ( with ceph ) or even kubernetes with, you guessed it - rook-ceph. Getting 10Gbit interface to work on ThinkCentre is possible, but not guaranteed to be reliable. This machine is perfect for such application and price point is not that terrible all things considered.
> Not sure about the CPU performance being much more powerful for some shit-stained NUCs found on ebay
The 10 GBit/s NUCs you find on eBay are enterprise-grade stuff: 10 Gbit/s hasn't really been a consumer thing. A used Fujitsu, Intel or Mellanox dual 10 Gbit/s bought on eBay isn't a "stained shit" that's "not guaranteed to be reliable". It's enterprise grade hardware.
I'd argue they're actually much better than some bargain-bin realtek cards, especially if you want raw performance.
They'll also probably work out of the box on whatever "server" distro you throw at them, which seems to be an issue with the machine in question.
However, GP was most likely talking about actual NUCs, not NICs. I mean, there wasn't a typo. The point probably being that the CPUs in some of those mini boxen are likely to be woefully undercooled, so performance may not be what you expect by just looking at the CPU model.
2x10G is the biggest selling point of this device. This can be very useful in certain use-cases, when you need a high speed interconnect with SSD-backed NAS, for example. Or between a Ceph-cluster nodes for the faster replication.
True. However, I've always noticed that ARM has less Linux support than x86, and the main benefits ARM is known for are typically performance/watt, running cooler, and less legacy support.
Since this server seems to have pretty average performance/watt and cooling, I can't really see much advantage to ARM here, at least for typical server use cases.
Unless you're doing ARM development, but I feel like a Pi 4/5 is better for basic development.
Linux support for ARM is inferior for end users of desktop 3rd party software. Everything else is provided by the repos. I doubt this person runs Signal or Spotify on those servers.
The performance is not average when compared with other ARM-based cheap computers, because it is high in comparison with them. It is also not average when compared to cheap Intel/AMD computers, because compared to such computers it is low. It could be called average only when averaging cheap ARM-based computers with cheap x86-64 based computers.
This computer uses 8 Cortex-A720 cores (and 4 little cores with negligible performance), which have a performance similar to the older Intel E-cores, i.e. Gracemont or Crestmont from Alder Lake, Raptor Lake or Meteor Lake. They are much slower than the recent Intel E-cores, i.e. Skymont or Darkmont, from Arrow Lake or Panther Lake.
So the performance of the whole CPU is similar to the 8-core Intel N300 (Alder Lake N) or Intel N350 (Twin Lake), which are found in various mini-PCs that are cheaper than this ARM computer.
Even so, the performance of this ARM CPU is many times greater than that of a Raspberry Pi and greater than of any cheaper ARM CPU. For greater performance, you must buy a more expensive smartphone, or a Qualcomm or Apple laptop or mini-PC, or a very expensive development computer from NVIDIA.
This is the only thing at a reasonable price with an Armv9.2-A CPU that is not a smartphone, but this Chinese CPU has various quirks.
An older but better ARM CPU with quadruple Cortex-A78 cores (Armv8.2-A ISA) is available for use in embedded computers from Qualcomm, rebranded from Snapdragon to Dragonwing. There are a few single-board computers of credit-card size with it, which are much faster than Raspberry Pi and the like.
Such SBCs are cheaper than the one from TFA and they are better for the purpose of software development.
The computer described in this article has the advantage of better I/O interfaces, the SoC has much more PCIe lanes, which allows the computer to have more and faster network interfaces.
If you want for an ARM computer to be a true high-throughput network server, then this one is the best choice. Nevertheless, for a true network server, a mini-PC with an Intel or AMD CPU will have a much, much better performance, at the same price or even at a lower price.
Using ARM is justifiable only for the purpose of software development, or if you want a smaller volume and a lower power consumption than achievable by a NUC-sized computer. For these purposes, one of the SBCs with Qualcomm QCM6490 is a better choice.
While a credit-card-sized SBC has only one Ethernet port, you can connect as many Ethernet interfaces as you desire to it (by using an USB hub and USB Ethernet interfaces), as long as the network throughput is not important and you just want to test some server software.
The Minisforum computer from the parent article has only 2 advantages for software development, the Armv9 ISA and being available with more memory, i.e. 32 GB or 64 GB, while the smaller ARM SBCs are available with 8, 12 or 16 GB.
Most people don't care about nominal difference in x86 vs arm. They care about cost, performance, efficiency, noise etc. Which applications run on the machine does matter.
The article never explained why the author wanted an ARM setup. I can only consider this a spiritual thing, just like how the author avoids Debian without providing any concrete explanations.
The usual reason to prefer ARM is efficiency, and the author's mention of replacing "power-hungry HPE towers" seeems to support that as a primary motivating factor.
True. But as detailed in the Jeff Geerling article that was shared here in the comments, it has (at least at the moment) a rather high idle power draw, which seems to negate that, especially over time.
This ARM computer has a much higher (3 to 4 times higher) idle power consumption than a mini-PC with an Intel or AMD CPU (e.g. an ASUS NUC), while having the same price and a much lower performance.
So in this case, the only valid reason to choose it is to have the ARM ISA for the purpose of software development.
This Chinese CPU is the only Armv9 CPU that is available in anything else than smartphones or expensive computers from Apple, Qualcomm or NVIDIA (or in even more expensive big servers). So there may be cases when it is desirable for software development, even if it has some quirks.
That is meaningful only if there is evidence to support that.
Mobile x86 processors used in mini PCs these days (as in 2026) are very competitive in terms of power efficiency. I wouldn't go for ARM just for that factor alone, especially without side-by-side comparisons of benchmarks.
> Most people don't care about nominal difference in x86 vs arm.
"Most people" aren't on HN, either.
The # of ARM servers at cloud providers are growing, but the ARM server options are severely lacking for most.
I, personally, would like to see more ARM growth (and I think we're heading that direction anyway... look at NVIDIA right now). Buying ARM servers that help push ARM software development forward is probably a good thing, IMO, from that POV.
> Most people don't care about nominal difference in x86 vs arm.
That's rubbish; even the people who don't care about ISA will care about stuff like power draw and software availability (although ironically arm seems distinctly worse in terms of power draw here).
But, I hope there are other people like me who will take a premium to avoid reading x86 core dumps, which is sort of like getting nails driven through your eyes. Yes, there's more software optimized for the chips; it is still bad code.
While logging serial number and some of the basic analytics stats might be attributed to stupidity, I tend to think that using a pretty advanced set of system commands and logging output consistently to log files is very sketchy.
One possible stupid-but-not-malicious explanation is that some anti-cheat company made a sketchy anti-cheat that includes server-side "is CheatEngine.exe running" code, and they're doing that via ps aux... and then this game player app was bullied by some game company into including this anti-cheat library to allow their game to run.
I hear this theory being claimed so much, but I don't see any real evidence for it; we have routers that you can monitor traffic on, we have microphone use indicators on mobile, and I would imagine it would be pretty clear if an app was uploading audio with even very basic monitoring tools. Correct me if I'm wrong, however.
I'm not denying that a lot of data is likely surreptitiously collected, but I'm talking microphone/camera in particular.
Most traffic is encrypted with HTTPS unless you can root every single device you own
we have microphone use indicators on mobile, and I would imagine it would be pretty clear if an app was uploading audio with even very basic monitoring tools.
Complicated smartphone OS, firmware, drivers might have bugs allow overrides of visual indicators.
That is fair. I do not think anyone could feasibly could detect/extract the exact data sent, because of HTTPS.
However I was more thinking of simple things, such as disabling anything that SHOULD be communicating with the Internet and seeing if any constant traffic persists.
Now of course, some very small (e.g plaintext) traffic might be almost undetectable, however that would suggest that most of the data would not be able to be transmitted due to size.
How confident or certain are you of what CSME or PSP or some code in TrustZone is doing? How certain are you that not a single piece of software on your machine, be it in the kernel, userland, drivers, is performing some type of surreptitious communication with CSME or PSP or program running in TrustZone?
Do you know for sure whether PSP or CSME has ever done DMA, or fingerprinted stack/heap allocation patterns and timing, or inspected the contents of your disk (after FDE was done being decrypted, of course), to evaluate whether common packet capture software is installed, or even whether it's currently running?
Detecting spyware is one thing. Detecting surreptitious nation-state spyware that behaves differently when it's being observed is a different challenge entirely.
In my case, I don't currently have any capture software on my main computer at all.
Our routers are Asus, and so I'm able to install tcpdump and log traffic directly without the source device itself knowing anything. This makes it really easy to monitor the traffic of any device, albeit not knowing exactly what it is being sent.
But it is true that I really can't know much more than what tcpdump shows.
I recall there were quite a few experiments where people use certain keywords heavily just to get closely related ads later on. I can totally relate my experience with it as well. Of course it is inconclusive - but if there is an incentive, management of big companies will venture into it. And chinese management is no different from western ones to that matter.
They don't pick the keywords uniformly randomly from a list of all keywords though. They think they randomly picked something that popped up in their mind, but those keywords are either
- stuff they saw online recently — ads or otherwise, which put the keywords in their mind
- or stuff they were already interested in recently
Not hard to imagine targeting algorithms picking up on either of these
You dont see those "coincidental" ads because your phone is listening to you, you see them because your freind showed interest in the product and theirs enough information to infer they talked to you about it. The good news is, your phone isn't listening to you without your consent. The bad news is, because it doesnt need to.
Admittedly, the 10G interfaces and fast RAM make up for some of it, but at least for a normal homelab setup, I can't think of an application needing RAM faster than even DDR3, especially at this power level.
reply