Maybe you were like me the other day, updating lots of machines all at once because a pair of massive PHP exploits leaked before they were supposed to be officially announced. Or maybe you saw what happened this morning, where a bunch of Linux security holes were disclosed on the wrong mailing list.

For mini PC users, issues like these are not trivial, especially if you’re running a lot of Docker images or running, say, a WordPress site on your stack. I personally spent a good part of Wednesday updating services I self-host to ensure they aren’t running outdated PHP versions. And, well … some were better about it than others. It took Ubuntu a couple of hours to update its apt repositories, but brew, an essential command line tool for Macs that is also popular with immutable distros like Bazzite, took nearly five hours to share the new PHP.

Which is to say that you were likely scrambling to ensure your setup was in working order this week, especially if you float across different versions of Linux. (Still haven’t done so? Go here to learn what you need to do next.)

It’s enough to make you give up, much like eMudhra did when its embarrassingly short TLS certificates were exposed to the broader internet. I saw some takes out there that defended the concept of software as a service in light of these security issues, notably from Beehiiv founder Tyler Denk. But I think that if you’re willing to self-host in your own home, using a mini PC or another method, you should do your homework and understand what you’re hosting. It can be really tempting to just keep up a bunch of random apps that you tried to self-host. It’s essential to periodically whittle down that list, and putting your stuff behind a VPN where possible is an important call. It’s one thing for a big PHP exploit to hit, but we live in an era when AI can discover security issues better than any human can.

In recent years, Ethernet speeds have continued to get faster, with speeds in certain specialized settings reaching 1.6 terabits, which is way faster than most demanding individual users are likely to ever need. But after years of one-gigabit connections, 10-gigabit Ethernet speeds are becoming increasingly common. The Mac Mini, for example, has offered the port as an upgrade option since 2018, and external Thunderbolt and USB-C adapters, while not cheap, are not uncommon.

You might be thinking to yourself, “but wait, my external internet connection tops out at one-gigabit speeds!” And you’d be right—the benefit of this port is likely not to get you the fastest possible internet speeds. Rather, the biggest benefit to a 10-gigabit connection is when you’re transferring a lot of files locally, for example when you’re sharing massive video files with your server, or in a local office. It’s a local networking boost, say if you’re sharing files between a file server or NAS.

(But if you’re wondering what it might be like to have a 10-gigabit internet connection, this blog post from Giles Thomas is enough to get you started.)

The problem for mini PC buyers, then, is that it’s often priced as a nice-to-have rather than a must-have, and you’ll therefore be paying a premium for the right to connect to your machines in an extra-peppy way. (That often leads to posts suggesting it’s overrated.) But if you only need a super-speedy connection sometimes, consider using a fast USB or Thunderbolt connection between two devices, a concept called IP Over Thunderbolt. That’ll give you many of the same benefits without the added cost and port complexities.

This is nearly two decades ago, but still recent enough that Zotac actually did a reveal on its YouTube channel.

If you were around in the late 2000s, you probably remember the all the excitement around netbooks, the small but usable computers that essentially put ASUS on the map. The Intel Atom-based chipsets on those could have potentially given mini PCs a bit of juice, but ultimately, they did not pop off in quite the same way the Eee PC did.

But at least one notable mini PC was able to leverage the energy around the netbook in desktop form, and it’s a company that’s still active today: Zotac. The company, which celebrated its 20th anniversary this year, first made an impression on reviewers with its “nettop” PC, the MAG ND01. While rocking just an Atom processor, it also came with Nvidia graphics, which helped it excel at gaming tasks of the era. (Turns out that the Nvidia Ion was a chipset made specifically for the netbook market. Fun! It’d probably add $500 to the cost of the machine in 2026.)

On HotHardware, reviewer Joel Hruska noted the machine was similar in size to the Nintendo Wii and, more amusingly, the Xbox 360 HD-DVD drive (talk about a relic of its time). Comparing it to an Asus AT3GC-I which sported the same Atom 330 chip but not the Nvidia graphics, he pointed out that the systems were more or less night-and-day performance-wise.

His summary:

Zotac’s MAG HD-ND01 is a great little system. It matches or exceeds the performance of Intel’s Atom 330 / 945G platform in virtually every test and it’s capable of playing some games at framerates you might actually want to play them at. Desktop performance under Windows 7 was surprisingly quick, the bundled wireless had no problem transferring data through several walls even though it lacks an external antenna, and the 2GB of RAM proved adequate to the tasks at hand. If there’s a performance shortcoming, it’s the hard drive. The modest Toshiba drive feels sprightly enough when you’re doing very basic sorts of tasks, but anything that hits the south bridge or accesses the drive can lag significantly.

Design-wise, it was closer in vibe to the early-2000s legacy-free PCs, though sporting some ports that made it better for home-theater use cases, like optical audio, which I need to write about sometime.

In 2015, Zotac put up a page on its website noting that the ND01 was an important machine in the company’s history. “Ever since, we’ve been pushing the limit of small form factor computing providing users with more and more functionality and features,” the company wrote.

These days, Zotac makes both mini PCs, graphics cards, and server gear, an interesting mix of product lines. But its legacy is perhaps most pronounced in the mini PC space, where it sort of predicted the Steam Machine-style PC decades ago. Though I can’t imagine anyone would pair an Atom processor with an Nvidia graphics chip in 2026.

If you’ve shopped for a high-end wireless router anytime within the last couple of years (an important feature for any Mini PC enthusiast), you probably know that TP-Link tends to hit closest to the price-performance sweet spot. But it’s long been dogged with questions about its Chinese origins, issues that have only become more prominent in the months since the FCC required pre-approval for new routers.

Last week, TP-Link got sued by five states that have suggested the router brand has failed to truly sever its ties to the Chinese market, based on past reporting from Bloomberg. (As one might guess, the states suing have a particular lean in their politics.)

The company, which has more or less dominated the home router market for the past few years, isn’t willing to cede ground to Netgear quite so easily. This week, it sent a letter to the FCC claiming that it was ready to play ball in the American market:

TP-Link is proposing to invest nearly $1 billion in new U.S.-based research and development and manufacturing capabilities. This investment would build upon TP-Link’s existing U.S. footprint, which includes a global headquarters in Irvine, Calif., and more than 700 U.S. employees. As part of this investment, TP-Link would begin assembling these Wi-Fi 8 devices in the United States by next year, and the devices will incorporate software and firmware developed, written, and stored in the United States by TP-Link’s U.S. engineering organization. All told, the contemplated investment would result in the creation of hundreds of high-paying research and development, manufacturing, and construction jobs for American workers over the next four years.

It’s not entirely clear whether TP-Link will be able to clear its name with China-skeptical politicians, but this much is true: This likely, at least temporarily, creates a gap in the market that competitors will rush through.

(By the way, in case you’re in need of a life-dominating hobby, did you know that you can actually buy Wi-Fi 7 router kits on Amazon? I guarantee you the parts in the Banana Pi BPI-R4 WiFi 7 won’t pass FCC scrutiny anytime soon.)

An emerging CPU instruction set architecture (ISA) that is positioned as something of a competitor to x86 and ARM processors. Unlike those architectures, it is fully “free and open,” meaning that anyone can make chips based on the architecture. At this point, its high end is nowhere near that of what x86 or ARM can do, which has limited its use cases mostly to embedded systems and some single-board computers. (It is leading to a lot of collaboration, including between the U.S. and China, which suggests if TP-Link wants to get politicians off its back, it should invest in RISC-V.) That said, there is at least one RISC-V mini PC out there, the Firefly AIBOX-K3, which sports a chip apparently capable of running 30-billion parameter AI models. Because everything has to be AI. Personally, my favorite use case for RISC-V is the Pinecil, a soldering iron from Pine64 that has a RISC-V brain.

For most end users, the Intel 386 line was dead as a doornail in 1994, with 486 machines extremely common and the Pentium starting to stream out to consumers.

But Intel had other plans for this chip, which it had introduced nine years earlier. In August of 1994, the company launched the 386EX, a processor that was intended to give its aging processor line a new home within the niche but still lucrative space of embedded systems. After all, conveyor belts need computing capabilities, too. But to get to where they needed to go, a few things needed to be cut out of these platforms, most notably a floppy disk. As a 1995 EDN piece noted, this wasn’t necessarily a problem for its use case. But you’ll have to do a lot of work to make it usable as a traditional computer:

If you’re building a “real” PC with the 386EX, you’ll need to add a few more peripheral devices, such as graphics, memory controllers, a keyboard, a real-time clock, a parallel port, and perhaps circuitry to generate the ISA bus signals if you plan to support some add-in cards.

That more or less ensured that these systems would rarely peek their heads out of the industrial doldrums, the kind of places where you might see an Arduino or Raspberry Pi today. But these chips nonetheless crept out into enthusiast contexts anyway.

One company particularly invested in the idea that these old 386 embedded chips could make good small computers was the German company Taskit, which sold two such devices, the MINI-PC and the 386EX-Card III, which was described as “one the smallest DOS computers on the market.” At 54mm by 96mm, it was only slightly larger than the standard dimensions of a Raspberry Pi.

These machines were not meant for plugging into an HDMI port and a PS2 keyboard. They didn’t have those ports. Rather, the assumption was that you would use it with a serial port through the console, rather than plugging in a monitor or keyboard. In other words, these were mini PCs for hardcore nerds and engineering types, rather than set-it-and-forget-it Mac Mini fans.

If you feel like getting your hands on a processor that was outdated upon its initial release, you can’t do much better than eBay, where such boards are out there, if not exactly common … or cheap. (Example: A 386EX-Card III listed on eBay as “rare” is currently selling for right around $1,000.) An alternative in the same vein that is likely going to be a bit friendlier for retro computing fans is the Vortex86DX, which is compatible with the 386 and tends to have more enthusiast interest. There’s also the more recent Pixel x86, which is built for retro gaming, but where’s the fun in that?

You must connect to a serial port to use a computer that’s this small and old, obviously.

If the news about the RTX Spark above made you feel a little sad about the state of the low end of the computer market, worry not. There are always options out there if you look hard enough.

One of those options comes in the form of thin clients, such as the Dell Wyse 5070, a circa-2018 machine that is getting fresh enthusiast attention as a cheap option. It’s still perfectly suited for some low-end computing tasks of the kind you might run in a homelab. As Patrick Campanale on HowToGeek notes, they’ve improved in value as the Raspberry Pi has crept up in price:

The problem is, while the Raspberry Pi used to be priced well, it simply isn’t anymore. Back in 2016 and 2019, a 1GB computer for $35 was a pretty great price. In 2026, you’d be hard-pressed to find a Pi 4B for $45 in stock anywhere, let alone $35.

The memory crisis has priced the Raspberry Pi out of the area it once dominated—affordable computers. At this point, a Raspberry Pi with a decent amount of power is the same cost as a mini PC without many upsides for the homelab. That’s why I’m starting to tell people to just look at thin clients for their homelabs.

(Even though it’s a weaksauce performer for an Intel machine, it’s still way faster than a the BCM2711 that comes in a Raspberry Pi 4. Plus, it benefits from the transcoding capabilities of Intel machines.)

Look, not everyone’s going to want a thin client hanging out in their closet, but in some contexts they can make a ton of sense.

A variant of mini PC, which actually came out in 2012 around the time of the first Intel NUC, that is intended as a ChromeOS-centric version of a thin client. These machines, especially earlier models, generally have 2GB or 4GB of RAM, just enough to do the basics. Later models, however, bumped up the specs accordingly. Don’t want to be stuck with ChromeOS on a used mini PC? Good news: These can have other operating systems installed on them, notably GalliumOS, but expect to open up the device.

One frequent knock against the various ARM plays on Windows is that Microsoft has traditionally been terrible at supporting ARM architecture, which limited its appeal and reach.

On Linux, this has rarely been a problem, thanks to prolific open-source contributors and thoughtfully targeted hardware. The words “Raspberry Pi” come to mind when bringing this architecture up, but if you want to go earlier, you might want to read up on the Corel Computing Netwinder.

If you think of Corel, you may think of graphics software or the classic word processor WordPerfect, which Corel acquired in 1996. But the company was one of the first big companies to invest in the Linux ecosystem in the big way, launching Corel Linux in 1999 … and abandoning it in 2001.

As a part of this initiative, Corel created a hardware offshoot, called Corel Computer, that was so obscure that it’s not even mentioned on the company’s Wikipedia page. But there is a CNET article, fortunately, and that article notes that the NetWinder was actually an attempt to broaden its fortunes away from the Windows ecosystem:

The NetWinder computers began as the platform Corel needed to sell its videoconferencing system, Lipson said. The Linux version of the Unix operating system and the StrongARM RISC chip provided a good avenue to make powerful yet inexpensive computers.

The firm had done other things to improve its lot in life, most notably reprogramming much of its software in Java, but this was easily the most ambitious thing.

These devices were said to be powerful but power-sipping, thanks to that ARM architecture. But just because it ran ARM doesn’t mean Intel wasn’t involved; in fact, Intel acquired the StrongARM brand in 1997.

The device was an innovative idea, but Corel quickly lost interest; by early 1999, it sold the machine to a firm called Hardware Canada Computing, a name that screams “generic.” Fortunately for them, the domain Rebel.com was available, giving the firm an opportunity to promote its Linux-friendly with images of James Dean. (Talk about on the nose.)

While this unusual machine didn’t last long, plenty of information is still out there about it online, thanks in part to the open-source backed resource NetWinder.org. The resource hasn’t been updated in over 20 years, but hey, at least the URLs work.

Mini PC fans and other self-hosting enthusiast types have not exactly had an event dedicated specifically to them, instead latching onto maker events, retro tech circles, and similar spaces.

All things that kind of touch on home labs and self-hosting, but never anything actually specifically targeted at the space.

But that’s about to change thanks to Jeff Geerling, a good dude who has been seeing some deserved success with his blog and YouTube channel over the past few years. Geerling is putting together Home Lab Fest, a St. Louis-area event that is already looking very promising for curious tinkerers. Beyond Geerling, pretty much every mini PC nerd you can think of on YouTube is making an appearance. (If all you watch on YouTube is ServeTheHome, Level1Techs, Craft Computing, and Alex Ziskind, do I have some good news for you.)

Boxed News is obviously new to this space, but this feels like our Super Bowl. Guess we gotta think about going to this; fortunately, it’s not hitting until September 12-14, 2027, so we have some time to think it over … though not too much, as early-bird tickets basically disappeared immediately.

As hinted at in the items directly above, there’s a constant demand for silent devices in the world of mini PCs. That’s especially true in industrial and similar settings, in part because a fan might be the only moving part in the entire system, and dropping a fan means it doesn’t need to be maintained.

So, how do you do it? Generally, this has meant making the entire case a giant heat sink, but in recent years, there’s been a spate of startups (including the aforementioned Frore Systems) attempting to make solid-state cooling devices. These essentially replace the spinning fan blades with vibration that can pull the heat out. That tech, while super-cool, is far from the mainstream at this point, so if you’re looking for a mini PC without a fan, you’re likely to see what is best described as a metal hedgehog with ports. Or, as it turns out, a Porcoolpine.

It’s one of many examples of how the heritage of mini PCs is generally embedded computing, rather than tiny shrinking boxes made by cool industrial designers like Jony Ive.

Thin clients are a kind of small computer that is designed as a waystation for a more powerful computer. It’s not a new concept—Sun Microsystems leaned into it back in the ’90s, and Larry Ellison arguably made his career on telling people about them, though it’s unknown if he actually helped sell any.

By the early 2000s, IT departments were realizing that it was a complete pain to have to manage computers at folks’ desks, and embraced thin clients as a way to help mitigate the pain.

But a company named ClearCube took things one step further, but putting the thin clients in the server room next to the server. This system, called the ClearCube C3, was very novel at the time, and essentially relied on putting complete systems on 5 1/4-inch “blades.” Instead, end users were given tiny desktop adapters that plugged into specialized Ethernet ports, allowing connectivity with keyboards, mice, USB components, and even monitors. (Yes, they did video over Ethernet. No, it apparently wasn’t very good. Yeah, I know. Wild.)

Given that the company’s product was getting compared against legacy-free systems like the Compaq iPaq, it really stood out as an unusual option, and one not without its problems, as a 2001 PC Mag review noted:

The blades slide into a cage, which can hold up to eight units and can be rack-mounted in a safe central location. For each blade in the rack, there are two RJ-45 ports: one for the C/Port, the other for an Ethernet connection. If a user shuts down his machine from the C/Port, a technician would have to turn it on from the blade.

Interestingly, the company is still around today, and pitches its connectivity approach as “zero client,” with products targeting government agencies and similarly high-end enterprise folks. Everyone needs a mini PC in a server room somewhere.

This week I’ve been traveling, and when I was on the road I wanted to take a quick gander at my server. I could have used my phone. Instead, I used an old e-reader I’ve been carrying around off and on for the past five years.

The Onyx Boox Nova3 Color, which I reviewed about five years ago on Tedium, is definitely showing its age at this point, but I’ve been able to cull a little more life out of it by making it a mini terminal on the go. Combined with a tethering-capable phone, it can be a decent tool for

One problem: Boox seemed to fight against my desire to install Termux at every turn. So I rooted it, which is pretty simple to do, as the device is not locked. (I have heard rumors to the contrary over the years, but to be clear: Yes, you can jailbreak it.)

I then set it up for a few niceties, including retheming the Termux interface for e-ink use, installing a lightweight e-ink Android launcher, and switching to the CleverKeys keyboard, which is capable of being set up to allow for a terminal-friendly keyboard layout. (It’s also the perfect layout for a six-inch tablet, so there’s that.)

Combined with Tailscale and some of the other terminal tricks I talked about last week, it makes for a lightweight way to, say, manaage an OpenCode instance remotely, or update your Docker containers while you’re riding the bus.

Problem is, not every e-reader is Android-capable—you probably want to lean on devices made by Chinese manufacturers like Boox or Bigme if you want that specific capability. It’s technically possible to install Android on an old Kindle, but it’s not exactly a friendly solution, with many jailbreaks requiring hardware mods. One emerging option: At least some of the uber-popular Xteink devices allow for adding custom firmware, with one known CrossPoint Reader port that adds a terminal, though that might be a little small for running on the fly.

There may be a time when you pick up an e-reader and you load up a terminal instead.

The ASUS Ascent QN10 mentioned above, sporting a Qualcomm Snapdragon X2, is noted for its fairly speedy neural processing unit (NPU), which works alongside its graphical processing unit (GPU). You might be wondering: Why should I care about a decent NPU?

Long story short: It’s effectively a very specialized portion of a modern processor that can blaze through very specific compute use cases, such as neural networks. As a 2012 research paper created by researchers at Microsoft Research and the University of Washington explained, NPUs expand the areas where neural networks can feasibly be used:

Rather than contributing a new design for neural network implementation, this paper presents a new technique for harnessing hardware neural networks in general-purpose computations. We show that using neural networks to replace regions of imperative code is feasible and profitable by experimenting with a variety of applications, including FFT, gaming, clustering, and vision algorithms. These applications do not belong to the class of modeling and prediction that typically use neural networks. For each application, we identify a single approximable function that dominates the program’s execution time. NPU acceleration provides 2.3x average whole-application speedup and 3.0x average energy savings for these benchmarks with average accuracy greater than 90% in all cases. Continuing work on NPU acceleration will provide a new class of accelerators—with implementation potential in both analog and digital domains—for emerging approximate applications.

While Microsoft helped develop the concept, Apple popularized the concept in mainstream technology, first adding it to its Apple A11 Bionic chip in 2017. If you owned an iPhone 8, you had an NPU in your hand all that time.

NPUs, which also found room in servers thanks to Google’s tensor processing unit setup, only hit mini PCs in the past few years, thanks to the M1 chip and, later, Intel and AMD processors. As AI becomes more of a thing, expect NPUs become a solid option for lower-end mini PCs down the road. The QN10 is sort of an early calling card on that front.

Trying to shrink down a mini PC to something small but actually still attractive is a bit of a tall order, as our piece on Tiny Computers highlighted last week.

But some did try, and one of the more interesting brands to do so was the Hush Mini-ITX, which was one of the first purpose-built computers built for that tiny-but-standardized form factor. The product of a company called Hush Technologies, it was not only built to be small, but silent, which was a fairly big deal in 2003.

Silence came with a cost, however. In a review with a similarly branded Tranquil PC unit, PC World noted that these machines were perfectly lightweight home-office machines, but struggled to do anything beyond that:

Despite their low benchmark scores, both machines proved quite capable of han¬ dling Web surfing, e-mail, and office tasks in informal testing. In the living room, each played DVD movies competently, and (working with my own add-in PCI TV tuner card) each successfully captured TV content at VHS quality. On the other hand, when I ripped a CD on them, both silent PCs took about twice as long as my PC (likely due to the limitations of their slim-line optical drives).

If you’re curious how a mini PC like this was marketed and built at the time of its release, This Studio, which helped design the form factor, has a whole page dedicated to the Hush machine on its website, complete with diagrams. It implies something that you might expect given its marketing: This was a mini PC intended for stereo systems.

Want to build a mini-ITX machine with this case? We found one on eBay, surprisingly.

How to SSH into your remote PC somewhat painlessly

If you’re a mini PC user, you likely will need to access your device “headless.”

And yes, the easiest way to do so is with a warm embrace of your terminal. You don’t need to necessarily use a terminal to run a mini PC; remote-access tools like RustDesk and Parsec make it more than possible to keep an eye on a device visually. But those come with overhead—they use extra resources and give you an extra desktop to manage.

Which means that if you want to get the most out of your mini PCs, you should leap into the terminal with both feet.

» A good starting point: If you use Microsoft’s Visual Studio Code, you can set up every machine you use for remote development with the Remote Development Extension Pack, which makes it possible to manage projects with ease. There are two main tactics: SSHing into a remote server, or using a VS Code Tunnel, which makes it possible to manage a given folder like it’s one of your VS Code projects, complete with extensions. One caveat: Microsoft doesn’t make a solution for mobile (it got a “not planned” status on its GitHub earlier this year), but others have attempted to fill the gap.

» A smart upgrade: But what if you want to get into your terminal without having to load VS Code on every device you own? That’s where Tailscale comes in. The networking tool, which does require a lightweight installer, is capable of connecting devices together within a contained network, which comes in handy for self-hosting, but its best trick might be the fact that it makes SSH authentication, a notably confusing process, painless. You can turn on or off remote terminal access on the fly through a web interface, a huge benefit if you only need to SSH into something every once in a while, and you can even access your servers without having to remember a password for every one. Tailscale is commercial, but has long offered a generous free plan. (If you want something that’s more integrated, the Terminus app uses a similar approach.)

» Getting fancy: The two above tools are of course great, but they come with an inherent problem: Every time you log into your terminal remotely, you have to start over from scratch. What can you do about it? Friends, this is why you should get to know tmux, a “terminal multiplexer” tool that makes it possible to run multiple terminals in a single process. But it’s capable of more than just that. It allows you to keep those processes running in the background remotely, which can then be accessed via SSH or other means. Trying to keep an eye on a complex task, like an AI run or a compilation process? Sticking it in tmux means that you can have it running on a mini PC while you keep an eye on it with a mobile terminal like Termux. There’s a learning curve, much as with any terminal-based tool, but it’s worth learning, given how powerful it is.

One more thought about learning curves: Don’t be afraid of them, especially if the tool works like other tools you might touch. You will be a better sysadmin for yourself if you can actually explain what the hell is going on with your damn terminal.

DP Alt Mode

USB-C is the greatest port ever, but some still call it the worst. And the confusing state of affairs over what the port is capable of doing is a big reason why. Nowhere perhaps is the confusion more acute than with the existence of DP Alt Mode, the primary way that we access video signals over USB.

Announced in 2014 by the Video Electronics Standards Association, it reflects the packet-based approach that both USB and DisplayPort are based on. In practice, it has made USB much more flexible, allowing more monitors to be plugged into a given machine. (Although your USB-C port and cable each needs to support the spec, and not all do.)

So you might be wondering: Why not HDMI Alt Mode? Yes, it exists, but as Pocket-Lint explained earlier this week:

Unlike HDMI, USB-C isn’t explicitly designed for audio and video transmission and uses an alternate mode to transmit them. It doesn’t support native video and audio transmission and instead uses HDMI or DisplayPort protocols. In reality, HDMI over USB-C hasn’t taken off, and most cables use the DisplayPort protocol, which supports up to 8K resolution at 60Hz and 100W of power transfer on a single cable.

This occasionally leads to the fun situation where your HDMI cables are shooting DisplayPort signals over an HDMI port. Exciting!

One would think if you named your company Tiny Computers, you’d be knee-deep in mini PCs. But you’d be wrong. The UK brand, which did try to launch an offshoot on the West Coast of the United States in the year 2000, was generally a big beige box brand, hitting the same market that Packard Bell did. (Sample ad language from its U.S. launch: “As American as steak & kidney pie.” No wonder we’re all using them today.)

But the brand did occasionally shrink its ambitions in a way that matched its name. In 2001, the company released the Takami, an AMD-based mini PC that aimed directly at the living room. Packing a DVD drive, it sold for £999 without a TV, but could be delivered with a 42-inch plasma TV for an eye-watering £6,999. (The Gateway Destination got there five years earlier, but who’s counting?)

Apparently this company was freaking massive with the Brits, despite the fact that the entire name of the company was mostly a lie.

Turns out, this device was a bet-the-company kind of thing. A Wall Street Journal report from around June 2001 notes that the device tested the company’s designers for eight months as it perfected the model. But even then it missed obvious things—like support for surround sound. When they added it in, it caused the machine to go haywire just before launch.

Folks were skeptical of the device, despite its marketing push, and the device ultimately sold fewer than 5,000 units—putting Tiny Computers in a tailspin that led the company to be bought out by its competitor Time Computers the next year.

Takami is Japanese for “excellent idea.” It should have been called Hayasugiru, which is Japanese for “too soon.”

Even low-end manufacturers are trying to make high-end AI boxes a thing

Of course, Nvidia wasn’t the only company trying to show off mini PCs with a little AI horsepower under the hood over in Berlin. Traditional mini PC manufacturers, like GMKtec, Acemagic, and Minisforum, are leaning on AMD’s Ryzen AI Max+ PRO 495, a.k.a. Gorgon Halo, to separate LLM tinkerers from their money.

These machines sport up to 192 gigs of memory, which is probably enough to run Deepseek v4 Flash locally. They’re complete overkill for all but the most dedicated of homelabbers. But they nonetheless represent an expansion of the sector that is even leading companies best-known for riding the low end of the market, like Chuwi, to go upscale.

(Chuwi doesn’t just sell low-end tech, but it did appear at The Verge recently with the headline “Is this $450 laptop from an unknown brand too good to be true?” So it has a rep.)

Which I guess raises the question: How big does the mini PC industry expect this market to be? At the moment, devices like the Framework Desktop are seen as niche or enthusiast plays. But if every company is making one of these, do they expect there to be a lot of people who just buy one of these to have around their house instead of paying OpenAI and Anthropic a monthly bill?

My prediction is that this market actually gets interesting in a few years, when the Ryzen AI Max+ machines become old news and can be found on the used market for $1,000 or less. That puts this otherwise exciting tech in the hands of more people with presumably more normal technology needs. But given that machines with the AI Ryzen AI Max+ 395 are selling for upwards of $3,000 on eBay, we might all be waiting for a while.

WTF is an M-Key?

They’re ubiquitous for SSDs these days, but not all modern M.2 devices are built the same way. See, the M.2 connector standard, first introduced in 2013 by the Peripheral Component Interconnect Special Interest Group, relies on a number of different “keys”—i.e. connector types—that are mostly physically incompatible with one another. For example, an E-key M.2 card, typically used for Wi-Fi or cellular connectivity, has a divider in the middle of its connector. The M-key, most common on modern-day NVMe-based SSD devices, has its divider on the right. But one potential bugaboo comes in the form of B-key devices, which have a divider on the left and are mostly used for SATA drives. Some of these SATA drives have what’s called a B+M key, making them work on either a B-key or M-key M.2 slot. Those may work in your modern-day mini PC, but in most cases you’re probably going to want an M-key drive, as NVMe drives are significantly faster than older SATA drives. But if you’re dumpster-diving, you should definitely be aware of the different types.

Hey! You got AMD graphics in my Intel NUC

Since we’re talking about mini PCs that combine the GPU and CPU power in a single chip, let’s bring up one of the more cursed entries in mini PC history. In 2018, Intel and AMD decided to actually collaborate on a couple of processors together, the i7-8809G and the i7-8705G, both sitting under the Kaby Lake code name. These processors included the usual Intel integrated graphics but also added a dedicated Radeon RX Vega M graphics chip into the mix.

While these chips appeared in some laptops, the main way this cursed collaboration reached the public was via the “Hades Canyon” NUC, a hefty gaming-targeted Intel mini PC with a glowing skull on the side. (It needed the skull, because if it didn’t have it, it would make the gamer-targeted box look like a router.)

It wasn’t the first Intel NUC in this form factor, but the two-companies-one-chip selling point gave the device a real air of novelty. Jeff from Craft Computing once reviewed this thing more than four years after it came out, and still found the device worthy of his time. In the pre-Intel ARC era, this was about as good as you could get as far as a GPU on package with an Intel laptop chip.

Plus, the combination meant that it was an excellent machine for folks who wanted to build a Hackintosh with a little bit more heft than a 2018 Mac Mini. The chipset was comparable to what you could get from a high-end 2018 MacBook Pro, with the RX Vega M more or less equal to the Radeon Pro 5500M that came with the high-end version of that machine. The CPU is showing its age in 2026, but graphics performance still holds its own comparable to more modern AMD iGPUs like the Radeon 890M.

It’s a machine I and lots of other people have looked longingly at over the years, and honestly, it still kind of feels like a decent buy today, nearly nine years after it was first announced. Packed with IO and performance, it likely chews through way more power than a modern-day mini PC, but it also doesn’t embarrass itself either. These things sell for between $300 and $500 on eBay if you’re curious.

Got a weird mini PC in your archives? Send a pic to us, and we might feature it here.

Nerds struggling to find the hard-to-come-by Mac Mini breathed somewhat of a sigh of relief last week after Apple announced the newest generation of the desktop machine and its larger cousin, the Mac Studio. Despite the brand new M6 chip under the hood of the base model (and the debut of the M5 Ultra chip), there’s plenty to grumble about here. There’s a pretty long wait for the actual machines, and a jump of $300 in a single generation is bound to sting. (Beyond the base model, consider getting a second mortgage to pay for it.)

But still, Apple is at least raising the price at a time when its desktop machines are drawing more eyeballs than ever. For months, facing the dual pressures of a RAM supply crunch and a level of demand well beyond prior high marks, something had to give … and yes, it was the price. While the hype around OpenClaw isn’t deafening anymore, the demand hasn’t gone away. (In case you’re wondering, Apple did not see the demand for agents pushing the machines into the stratosphere, which is one reason why desktop Macs have been so hard to come by of late.)

All in all, it’s a spec bump, but a pretty good one—a perfect, pat victory lap for John Ternus as he takes over for Tim Cook as Apple CEO starting today.

But at least one small detail hidden in the announcement might have been a legit accident. Per MacRumors, the Cupertino crew jumped the gun by suggesting the SSD in the Mac Studio utilized PCIe gen 6, which has yet to appear in a consumer product.

And hey, maybe the release of new Mac Minis will finally get those absurdly overpriced eBay listings to calm the heck down. Or maybe OpenAI is just going to keep buying the machines and artificially raising prices for the rest of the market.

I will wave my arms around and point out, hey, there’s a whole rest of the industry out there. But the Mini nonetheless remains at the top of the heap.

If you are even possibly thinking of getting an external GPU for your higher-end mini PC, you have probably heard of OCuLink and Thunderbolt. (For the layperson: OCuLink is faster and extremely stable, but not hot-pluggable; Thunderbolt is more common and more versatile, but slower.) But some recent mini PCs, notably a high-profile entrant from GPD, sport a Mini Cool Edge IO® (MCIO) port. WTF is that? Well, like OCuLink, MCIO essentially a server interconnect technology that found a use case in the consumer realm thanks to the emergence of the eGPU.

Developed by Amphenol, MCIO makes it possible to get near-native PCIe speeds from an external device. OCuLink does, too, but MCIO has almost zero drop-off. So what’s the downside? On top of its rarity outside the server room, the port itself is also a bit of a beefy boy, looking kind of like the second coming of PCMCIA compared to OCuLink’s HDMI-esque connector. Put simply, if OCuLink is niche, MCIO is even further down the rabbit hole.

Odds are you need to be really into eGPUs for MCIO to matter. But, it could be a bigger deal down the road.

The Mac Mini might have revolutionized the mini PC by fitting the whole thing into a flat rectangular case way back in 2005, but that meant there was a ton of room for catch-up after the fact. One of the earliest to try was AOpen, which announced its first mini PC back in 2005. Sporting an Intel Pentium M, a processor that would appear in the original Apple TV set-top box the next year, the micro aimed for the same market that the Intel NUC would develop half a decade later.

Reviewers were enamored. While raising concerns that Apple might attempt to sue AOpen, Maximum PC wrote that the machine was “an amazing balance of miniaturization and user-friendliness.” And unlike the original Mac Mini, which infamously required a putty knife to open, mere mortals could make their way into the AOpen.

“We expected the Mini to be horrible to wrench on, but it’s an easy unit to crack open,” the outlet stated.

By the way, AOpen is still at it today, by the way, mostly focused on embedded systems. And its early DNA appears to have seeped through; its modern-day AOpen Ace Mini continues to rock the distinctive on-button-in-the-middle design of its first mini PC.

Got a weird mini PC in your archives? Send a pic to us, and we might feature it here.