The performance on this machine is similar to the Snapdragon 8cx from Microsoft’s first big Windows on ARM push a few years ago (so nowhere near even the 4 year old Apple M1) while the idle power draw is similar to an x86 computer. If they’re not targeting either performance or power efficiency, I’m not sure who would be interested in this computer. I’d say that if you can tolerate Mac OS X for your use case, a used Mac Mini would be a far better choice.
The one niche they could hit I think is Arm devs who need an Arm machine, want to run Linux (and not rely on Asahi, which still only runs on M1/M2), and want some hardware expansion (PCIe, M.2, dual 10 GbE, etc.).
That's a lot smaller than the general homelab niche, or those who want a small, efficient, quiet computer for some purpose.
As I mentioned in the post, the M4 mini is only $50-100 more (though RAM upgrades quickly negate the price comparison past the base model), and it's many times more efficient—and powerful.
And while you can't do everything with Thunderbolt... there are some halfway-decent external PCIe docks you can use with some devices on macOS now.
I don't understand this niche. Linux does not need to run on an ARM machine for almost anything. Linux has very good cross platform abilities and so you can build and test on whatever CPU (presumably x86, but any CPU will do), and then when you are confident that everything works cross compile for ARM and things will almost always work. You also have qemu which is slow but plenty good for running some code where this matters.
You do need a little testing because ARM is different, but the odds that are are doing something where it matters are low. I've been doing the above for years and only found two things. First was a bug in GCC (already fixed in a newer version by the time I traced it down). Second was x86 has a strong stronger memory model for sharing data between threads - hopefully you are not doing that (I only hit it because I maintain our cross thread message system).
You also can't test anything that uses GPIO type things - but this computer has different setup and so you couldn't with this anyway. (and you should abstract your GPIO for testing anyway so this because a small test case when you do switch to real hardware)
I think I'm probably in the target market for this. I have a board using Linux on an Arm v9.2 chip. It'd be nice to run unit tests with exactly the same binaries and compiler settings. The actual boards are a bit expensive and unavailable for CI purposes (vs HIL). Graviton isn't v9.2. QEMU doesn't support v9.2. This does.
Probably won't end up buying this, but it's not hard to imagine a small number of other people would find it useful.
This, basically - if you target embedded, automotive, mobile (Android, perhaps), then having good native machines supporting the latest Armv9 features is helpful, compared to emulating it on another machine, or running stuff in the cloud.
Like I said, a niche, but not one that's nonexistent. I know Minisforum was hyping the Android angle a bit, though in my small amount of mobile app dev, I was happy enough doing that on my Mac.
Ironically, the most Linux on ARM I have experience with is from a VM on my Mac. It wouldn’t be perfect , but it’s pretty good for working with ARM. You aren’t emulating the CPU architecture, but you are running under virtualization.
I’m just thinking that the Mac Mini might also be better for that usecase, even with the virtualization in place. If you need to support a specific processor, you probably want to work with that exact processor. But if you want to use any ARM, a virtualized Linux running on a Mac isn’t a bad option.
Using UTM I managed to get a headless server running on the Mac mini that behaved better than the mini did. Eg mounted file servers actually mounted at boot reliably and container startup was controlled (maybe it’s me but desktop docker… ew).
The downside is that it eats into power usage. Even then, I’m averaging just 6.4w over 30 days.
portable code is starting to become valued in automotive. In the 8 bit era rewriting for each new modle was a good choice but there days code is too complex and so we need our code to survive longer than the cpus we start with. Often the cpus a car starts with are out of porduction before the model run is done. Car volumes are not big enough to keep a fab open.
Besides TSO like you mentioned, the SIMD implementation is different (SSE/AVX vs. NEON) which is important if you do anything that uses SIMD intrinsics directly. If you're deploying to ARM servers, it's also nice to be able to test the exact same binaries you'll be running on the server.
Yeah Asahi makes this machine even more of a niche product. Even if you're stuck on the M2 Mac Mini, that still gets you ~2x the single core performance and ~40% more multi-core performance than the MS-R1, with much better power efficiency.
> That's a lot smaller than the general homelab niche, or those who want a small, efficient, quiet computer for some purpose.
The homelab niche always surprises me for how many people are willing and able to buy hardware they don't really need.
I'm sure it's still a small niche, but I wouldn't be surprised if minisforum was selling a lot of devices to homelab builders looking for the next toy to add to their collection.
If you can't tolerate macOS, you can get a used M1/M2 which will run Linux really well.
That's a challenge this machine has, it has to compete with things like a used 64GB M1 Max Mac Studio for $900 when the equivalent config is $700. The M1 Max would be wildly faster in everything.
I consider myself relatively adventurous with ARM SBCs, but I don't see the draw for this box. $500 for the 32GB of RAM version is a lot to ask, especially with no path to upgrade the RAM and coming from a company known to have mediocre quality control and difficult warranty support.
I always appreciate Jeff Geerling's willingness to do in-depth reviews and comparisons. In this case I would have liked to see a cheap x86-64 option in the ranks, though. I understand that it's supposed to be about ARM and low power, but with this box drawing 17W at idle and coming with a $500 price tag for 32GB RAM and no storage, we might as well start comparing to x86 options in the same price and power envelope.
I got a defective MS-01 and the return process was smooth and they paid for everything. I got a full refund even after the Amazon refund window expired.
I have a MS-A1 and a MS-A2 now and both have been serving massive web traffic for the past 8 months with 0 issues.
I think minisforum products are generally decent, but there are things that give me pauses.
I had a mini PC that would sometimes briefly show black screens followed by "AMD driver timeout" issues. I emailed their support, who instructed me to play with a few BIOS settings (slightly weird), and eventually asked me to start a warranty replacement. The emails were not tracked by any ticket system, just as regular emails, and the shipping address was a residential address in California. All of this feels slightly unprofessional, and the residential address thing is almost alarming. But in the end, the issue got addressed, so whatever.
The overall service definitely cannot possibly match Intel with their NUCs or Apple products, not a surprise. But the actual machines are ok.
If your mini PC died within warranty period, I don't see why you cannot do warranty returns. Otherwise, you just had bad luck, just like what could happen with many other devices from other manufacturers.
I can't speak for the Intel ones but I hear the AMD Zen 4 and later ones have been solid. I have a UM780 XTX that I've been using in a homelab setup and it's been rock solid. Zero issues with Linux (Arch btw), zero hardware issues after nearly 2 years of 24x7 operation.
These Chinese mini PCs are generally quite unreliable.
I have one Minisforum that had the HDMI glitch out periodically and another that would reboot on high MEM load. And the stock SSD died without any warning.
I relegated them to a NAS role and at low loads they’re ok.
Yeah I don't mind a ~250ish gamble on aliexpress but 500+ minipcs is above my risk tolerance. At that level I'd rather DIY a desktop sized build where I can troubleshoot individual components
I really like Minisforum in general and the current generation of AMD-powered Mini PCs. I have one as a media PC and I replaced my dad's desktop with one. They are cheap, fast and silent.
I was thinking of getting a high end AMD one as a replacement for my daughter's aging gaming PC, although as of this week I am thinking of waiting for Steam Machine.
This ARM machine seems a bit slow (I would have suggested a new Qualcomm CPU, some of them are crazy fast) but it is nice to see a major Mini PC manufacturer getting into the ARM space.
Having bought other systems in the MS-line, these boxes are heavy on cool homelab specs but aren't all that reliable and dealing with Minisforum for support is abysmal. I ended up moving back to old enterprise hardware (TinyMiniMicro or rackmounted gear). I don't recommend purchasing these systems if you're looking for anything more than a novelty; they're not any better than buying one of the random AliExpress mini PCs.
I still bought my 795s7 (a cheap 16-core MoDT (mobile on desktop) zen4 box) anyways, but accepting that I would probably never get bios updates ever was a hard pill to swallow. I haven't had to deal with support for anything defective, but support as in available bios updates being so so so spotty is really unfortunate.
AMD is talking about replacing closed AGESA BIOS with open OpenSIL bios some day, and maybe perhaps possibly it means end users get some chance to maintain & upgrade bios themselves, eventually, possibly. Given how some vendors seem uninterested in doing the work themselves, this sliver of a hope would be nice to see happen.
I have the BD795M, the support is non existent and the BIOS updates are questionable. There exists a BIOS update but I've read it breaks more than it fixes. At current the temperature control of the fans is just abysmal. I can live with that for now.
It would be great to have a full Snapdragon X Elite-based box, with working Wi-Fi drivers.
That said, it feels as if the fragmentation in the non-Windows space ends up being worse for non-Intel/AMD platforms, both commercially and from a devrel perspective. Qualcomm and Apple still have the best arm64 platforms above a Raspberry Pi.
This was my experience looking for an ARM Linux machine. The Apple M-series processors were so far ahead of anything else currently available on the market for the price. And the Ampere stuff is in an Epyc class of price with not an Epyc class of performance.
I think that is a result of this being their first entry into the ARM market. IT takes a while to get all the hardware and drivers in sync. Hopefully they don't give up and also upgrade to a better CPU.
There has got to be some firmware bugs causing high power usage, but also the CIX CPU is fabbed on TSMC 6 nm process which is a couple generations old.
And note that Radxa launched their O6 with the same SoC earlier this year, and they haven't brought down idle power consumption either.
There's also the Orange Pi 6 Plus, and the Radxa O6N with the same SoC, they would all benefit from the power issues on this chip being resolved. Not sure if it'll happen.
I bought an MS-01 (the earlier Intel version of this) for homelab use and I don't know what was wrong with it, but Debian just refused to reboot on it. It would boot, but not reboot, freezing up every time.
I replaced it with an Asus NUC, which came with a non-functional BIOS, but was eventually coerced into working after forcibly flashing it.
That power usage is abysmal. I can’t see that it states how the draw was measured, but on a mini my average usage is 6.4w over a month, and that includes all usage. I moved from a nuc 9 which was using about 70w.
I don't like the word "homelab". Some people have a lab at home (computing, optics, chemistry, wherever) - that's fine. But this is just... a computer.
The performance on this machine is similar to the Snapdragon 8cx from Microsoft’s first big Windows on ARM push a few years ago (so nowhere near even the 4 year old Apple M1) while the idle power draw is similar to an x86 computer. If they’re not targeting either performance or power efficiency, I’m not sure who would be interested in this computer. I’d say that if you can tolerate Mac OS X for your use case, a used Mac Mini would be a far better choice.
The one niche they could hit I think is Arm devs who need an Arm machine, want to run Linux (and not rely on Asahi, which still only runs on M1/M2), and want some hardware expansion (PCIe, M.2, dual 10 GbE, etc.).
That's a lot smaller than the general homelab niche, or those who want a small, efficient, quiet computer for some purpose.
As I mentioned in the post, the M4 mini is only $50-100 more (though RAM upgrades quickly negate the price comparison past the base model), and it's many times more efficient—and powerful.
And while you can't do everything with Thunderbolt... there are some halfway-decent external PCIe docks you can use with some devices on macOS now.
I don't understand this niche. Linux does not need to run on an ARM machine for almost anything. Linux has very good cross platform abilities and so you can build and test on whatever CPU (presumably x86, but any CPU will do), and then when you are confident that everything works cross compile for ARM and things will almost always work. You also have qemu which is slow but plenty good for running some code where this matters.
You do need a little testing because ARM is different, but the odds that are are doing something where it matters are low. I've been doing the above for years and only found two things. First was a bug in GCC (already fixed in a newer version by the time I traced it down). Second was x86 has a strong stronger memory model for sharing data between threads - hopefully you are not doing that (I only hit it because I maintain our cross thread message system).
You also can't test anything that uses GPIO type things - but this computer has different setup and so you couldn't with this anyway. (and you should abstract your GPIO for testing anyway so this because a small test case when you do switch to real hardware)
I think I'm probably in the target market for this. I have a board using Linux on an Arm v9.2 chip. It'd be nice to run unit tests with exactly the same binaries and compiler settings. The actual boards are a bit expensive and unavailable for CI purposes (vs HIL). Graviton isn't v9.2. QEMU doesn't support v9.2. This does.
Probably won't end up buying this, but it's not hard to imagine a small number of other people would find it useful.
This, basically - if you target embedded, automotive, mobile (Android, perhaps), then having good native machines supporting the latest Armv9 features is helpful, compared to emulating it on another machine, or running stuff in the cloud.
Like I said, a niche, but not one that's nonexistent. I know Minisforum was hyping the Android angle a bit, though in my small amount of mobile app dev, I was happy enough doing that on my Mac.
Ironically, the most Linux on ARM I have experience with is from a VM on my Mac. It wouldn’t be perfect , but it’s pretty good for working with ARM. You aren’t emulating the CPU architecture, but you are running under virtualization.
I’m just thinking that the Mac Mini might also be better for that usecase, even with the virtualization in place. If you need to support a specific processor, you probably want to work with that exact processor. But if you want to use any ARM, a virtualized Linux running on a Mac isn’t a bad option.
Using UTM I managed to get a headless server running on the Mac mini that behaved better than the mini did. Eg mounted file servers actually mounted at boot reliably and container startup was controlled (maybe it’s me but desktop docker… ew).
The downside is that it eats into power usage. Even then, I’m averaging just 6.4w over 30 days.
portable code is starting to become valued in automotive. In the 8 bit era rewriting for each new modle was a good choice but there days code is too complex and so we need our code to survive longer than the cpus we start with. Often the cpus a car starts with are out of porduction before the model run is done. Car volumes are not big enough to keep a fab open.
Besides TSO like you mentioned, the SIMD implementation is different (SSE/AVX vs. NEON) which is important if you do anything that uses SIMD intrinsics directly. If you're deploying to ARM servers, it's also nice to be able to test the exact same binaries you'll be running on the server.
Yeah Asahi makes this machine even more of a niche product. Even if you're stuck on the M2 Mac Mini, that still gets you ~2x the single core performance and ~40% more multi-core performance than the MS-R1, with much better power efficiency.
> That's a lot smaller than the general homelab niche, or those who want a small, efficient, quiet computer for some purpose.
The homelab niche always surprises me for how many people are willing and able to buy hardware they don't really need.
I'm sure it's still a small niche, but I wouldn't be surprised if minisforum was selling a lot of devices to homelab builders looking for the next toy to add to their collection.
If you can't tolerate macOS, you can get a used M1/M2 which will run Linux really well.
That's a challenge this machine has, it has to compete with things like a used 64GB M1 Max Mac Studio for $900 when the equivalent config is $700. The M1 Max would be wildly faster in everything.
Doesn't asahi run on the Mac mini? Probably better than it does on a laptop, no peripherals to support
Thunderbolt and DP Alt mode aren’t supported, and there are issues with power management (sleep/wake).
OK, but does that matter much for a home lab?
It depends, you decide.
Once I looked at the price tag I quickly arrived at the same conclusion. It's a neat product, but I can get so much more computer for less money.
I consider myself relatively adventurous with ARM SBCs, but I don't see the draw for this box. $500 for the 32GB of RAM version is a lot to ask, especially with no path to upgrade the RAM and coming from a company known to have mediocre quality control and difficult warranty support.
I always appreciate Jeff Geerling's willingness to do in-depth reviews and comparisons. In this case I would have liked to see a cheap x86-64 option in the ranks, though. I understand that it's supposed to be about ARM and low power, but with this box drawing 17W at idle and coming with a $500 price tag for 32GB RAM and no storage, we might as well start comparing to x86 options in the same price and power envelope.
My minisforum mini PC is dead and useless. I will never, ever recommend anyone to buy a minisforum again.
Please do not buy a minisforum.
And it wasn't just me. https://old.reddit.com/r/MiniPCs/comments/1ocvjby/minisforum...
I got a defective MS-01 and the return process was smooth and they paid for everything. I got a full refund even after the Amazon refund window expired.
I have a MS-A1 and a MS-A2 now and both have been serving massive web traffic for the past 8 months with 0 issues.
I think minisforum products are generally decent, but there are things that give me pauses.
I had a mini PC that would sometimes briefly show black screens followed by "AMD driver timeout" issues. I emailed their support, who instructed me to play with a few BIOS settings (slightly weird), and eventually asked me to start a warranty replacement. The emails were not tracked by any ticket system, just as regular emails, and the shipping address was a residential address in California. All of this feels slightly unprofessional, and the residential address thing is almost alarming. But in the end, the issue got addressed, so whatever.
The overall service definitely cannot possibly match Intel with their NUCs or Apple products, not a surprise. But the actual machines are ok.
If your mini PC died within warranty period, I don't see why you cannot do warranty returns. Otherwise, you just had bad luck, just like what could happen with many other devices from other manufacturers.
I can't speak for the Intel ones but I hear the AMD Zen 4 and later ones have been solid. I have a UM780 XTX that I've been using in a homelab setup and it's been rock solid. Zero issues with Linux (Arch btw), zero hardware issues after nearly 2 years of 24x7 operation.
I've owned a couple and Minisforum has generally been amazing.
Same, I have a fairly positive experience with minisforum.
These Chinese mini PCs are generally quite unreliable.
I have one Minisforum that had the HDMI glitch out periodically and another that would reboot on high MEM load. And the stock SSD died without any warning.
I relegated them to a NAS role and at low loads they’re ok.
Yeah I don't mind a ~250ish gamble on aliexpress but 500+ minipcs is above my risk tolerance. At that level I'd rather DIY a desktop sized build where I can troubleshoot individual components
Same here. Did yours have a fan or was it fanless?
It has a fan. I bought it from Amazon and it has been over a year but when I emailed them, I got no response at all.
I really like Minisforum in general and the current generation of AMD-powered Mini PCs. I have one as a media PC and I replaced my dad's desktop with one. They are cheap, fast and silent.
I was thinking of getting a high end AMD one as a replacement for my daughter's aging gaming PC, although as of this week I am thinking of waiting for Steam Machine.
This ARM machine seems a bit slow (I would have suggested a new Qualcomm CPU, some of them are crazy fast) but it is nice to see a major Mini PC manufacturer getting into the ARM space.
Having bought other systems in the MS-line, these boxes are heavy on cool homelab specs but aren't all that reliable and dealing with Minisforum for support is abysmal. I ended up moving back to old enterprise hardware (TinyMiniMicro or rackmounted gear). I don't recommend purchasing these systems if you're looking for anything more than a novelty; they're not any better than buying one of the random AliExpress mini PCs.
I still bought my 795s7 (a cheap 16-core MoDT (mobile on desktop) zen4 box) anyways, but accepting that I would probably never get bios updates ever was a hard pill to swallow. I haven't had to deal with support for anything defective, but support as in available bios updates being so so so spotty is really unfortunate.
AMD is talking about replacing closed AGESA BIOS with open OpenSIL bios some day, and maybe perhaps possibly it means end users get some chance to maintain & upgrade bios themselves, eventually, possibly. Given how some vendors seem uninterested in doing the work themselves, this sliver of a hope would be nice to see happen.
I have the BD795M, the support is non existent and the BIOS updates are questionable. There exists a BIOS update but I've read it breaks more than it fixes. At current the temperature control of the fans is just abysmal. I can live with that for now.
How is the 795s7? What do you put it to work on?
Just sprung for one at a good price.
I have the BD795M with an RTX 3090. It's a good desktop and gaming machine if you're looking for a compact build.
Support is abysmal
It would be great to have a full Snapdragon X Elite-based box, with working Wi-Fi drivers.
That said, it feels as if the fragmentation in the non-Windows space ends up being worse for non-Intel/AMD platforms, both commercially and from a devrel perspective. Qualcomm and Apple still have the best arm64 platforms above a Raspberry Pi.
This was my experience looking for an ARM Linux machine. The Apple M-series processors were so far ahead of anything else currently available on the market for the price. And the Ampere stuff is in an Epyc class of price with not an Epyc class of performance.
More than anything, reading this review acts as an ad for the Mac Mini M4. $600, with performance and efficiency well beyond the other options.
the performance and power usage is pretty bad on it. for $50 more you can get a Mac mini M4 that blows it out of the water.
I think that is a result of this being their first entry into the ARM market. IT takes a while to get all the hardware and drivers in sync. Hopefully they don't give up and also upgrade to a better CPU.
There has got to be some firmware bugs causing high power usage, but also the CIX CPU is fabbed on TSMC 6 nm process which is a couple generations old.
Supposedly that could be fixable with a bios/firmware update… but there’s no guarantee that will ever happen
And note that Radxa launched their O6 with the same SoC earlier this year, and they haven't brought down idle power consumption either.
There's also the Orange Pi 6 Plus, and the Radxa O6N with the same SoC, they would all benefit from the power issues on this chip being resolved. Not sure if it'll happen.
I bought an MS-01 (the earlier Intel version of this) for homelab use and I don't know what was wrong with it, but Debian just refused to reboot on it. It would boot, but not reboot, freezing up every time.
I replaced it with an Asus NUC, which came with a non-functional BIOS, but was eventually coerced into working after forcibly flashing it.
I haven't had a lot of luck with mini PCs.
Running Linux directly on ARM is the dream, but this looks slow. So for now I'm stuck with UTM VM's on the Mac.
DGX Spark but don't ask the price.
That power usage is abysmal. I can’t see that it states how the draw was measured, but on a mini my average usage is 6.4w over a month, and that includes all usage. I moved from a nuc 9 which was using about 70w.
I don't like the word "homelab". Some people have a lab at home (computing, optics, chemistry, wherever) - that's fine. But this is just... a computer.