**Nobody builds a MiSTer to play games. At least not at the start.**
I lost an entire Saturday because of an SD card. It still annoys me to think about it.
I had the board assembled, the fan purring, the MiSTer menu showing up on the monitor, and suddenly the Linux running inside decided the filesystem was dirty. I had pulled the plug from the wall, like you do with a 1992 console, and that's what broke it. A 1992 console doesn't have an operating system writing to disk when you're not looking. The MiSTer does. It's the first thing nobody tells you when they sell you the idea that FPGA is "real hardware": the thing living next to the real hardware is a small Linux computer, and that computer has opinions.
I'll explain what this is for anyone arriving now, and then I'll be tedious about the part that matters. The MiSTer is an open project that grew around a Terasic development board, the DE10-Nano. Inside is an Intel Cyclone V chip that joins two worlds. On one side, two ARM cores run Linux and handle the menu, the files, the commands and whatever you see on screen when you press the menu button. On the other, the FPGA fabric, where someone has rewritten in logic the circuit of a Mega Drive, an Amiga, a Capcom arcade board. The FPGA doesn't interpret Mega Drive instructions. It gets reconfigured to behave like a Mega Drive. The difference from a software emulator is real and not just marketing, but we'll get to that, because the marketing exists and it eats a lot of people alive.
What interests me here, and what I think almost every text on the subject handles badly, is the community. The board is a board. You can buy it, and for a while buying it was a problem, but on its own it's an expensive paperweight. The people who wrote the cores, and keep writing them, are a small group working mostly for free, from incomplete documentation, from chips someone opened up and photographed under a microscope, from test ROMs made by obsessive people in the nineties, and from pure stubbornness. Sorgelig started what became the core of the project, building on the MiST that already existed, and around him grew a GitHub organization, a forum, a Discord server where at three in the morning someone is always awake arguing about VDP behavior. Everything else, the pretty case, the YouTube video, the I/O board, is an accessory to this.
This changes how you should look at the thing. You're not buying a product. You're entering a dependency on other people's work that nobody promised you.
Putting a MiSTer together, in practice, is a sequence of small purchases and small disappointments. The DE10-Nano cost around 130 dollars in the good times, and then the 2021 chip shortage came and the board vanished from distributors for months, with resellers asking double and a bit more. On top of the board goes the SDRAM board, which is where a lot of people trip first. Linux and the core share the board's DDR3, and that memory, for anyone trying to imitate 1980s hardware timing, has too much latency and unpredictable behavior. Cores that need fast, deterministic memory use the SDRAM module that plugs into the expansion connector. There are 32, 64 and 128 megabyte versions, and the classic mistake is buying the 32 because it's cheaper and discovering, three weeks later, that the Neo Geo or Saturn core wants the bigger one, and the N64 one wants more than you ever thought a 1996 console could ask for. I did exactly this. I bought the SDRAM from a seller in Hong Kong, waited about five weeks, and when it arrived there was a notice from the Portuguese post office asking for 23% VAT plus a customs handling fee that felt invented on the spot. I paid. Of course I paid.
Then comes the analog I/O board, if you want RGB out to a CRT, the USB hub that isn't just any hub, the case, the fan, the power supply that has to handle real current and not some phone charger forgotten in a drawer. In one of my attempts I burned a pad off the USB connector because I insisted on soldering a wire with the iron at 400 degrees, convinced I knew what I was doing. I didn't. The smell of rosin stayed in the room for three days.
I remember, with no relation to any of this, the downstairs neighbor's Spectrum, with that rubber keyboard, and sitting on the floor of his living room watching an older guy write a BASIC program that drew a circle. A circle. I stared for half an hour. I don't know why this came to mind now, and I'm not going to try to tie it to the rest.
The question of accuracy, though, does need to be discussed with some care, and this is where the community splits most without admitting it. When someone says a core is "cycle accurate", they mean the original circuit's behavior was reproduced down to the clock cycle, including the quirks, the silicon bugs that programmers of the time exploited. That exists in some cores. In many others, what exists is a very good functional reimplementation, built from documentation and test ROMs, that behaves like the original in almost everything and disagrees with it in obscure corners. There is extraordinary work done by people who started from serious reverse engineering, people who opened chips, photographed the silicon and rebuilt the logic gate by gate, and some reimplementations of the Mega Drive sound chip benefited directly from that effort. But the label gets thrown around too easily, in forums and in videos, and I think that's dishonest, even when the intention is good.
Here comes my less popular opinion. For the vast majority of people, and honestly I think that includes me for a large part of the time, the difference between a good software emulator and an FPGA core can't be heard or seen. A well-written emulator, on a PC today, handles maybe 96% of use cases without anyone complaining. The rest is where the FPGA earns its money, and the rest is very small.
Which, of course, is exactly where this community lives. I've contradicted myself already, fine. But think about it: whoever builds a MiSTer isn't someone who wants to play Sonic. It's someone who wants to see whether the raster effect in a Mega Drive demo everyone uses to test accuracy, Overdrive 2, does what it does on the original. It's someone who spent their youth working out why Amiga music sounded the way it did, with that audio filter the hardware applied and no lazy emulator replicated. It's the person who knows by heart that the Amiga's Copper allowed per-scanline color tricks that demo coders exploited down to the last cycle. For that person the 4% is everything. And I get it, even if I laugh at it.
Then there's latency, the argument everyone uses and few people measure. The chain is long. The controller button, the USB, Linux, the core, the video scaler, the screen. Each link adds something. In a file called MiSTer.ini, which lives at the root of the card, there are lines that touch this directly. I use vsync_adjust=2 for low latency mode, which syncs the output with the source instead of letting the scaler decide on its own, and hdmi_game_mode=1 to ask the display to turn off image processing. Without it, the monitor treats the output like a movie. With it, and with a monitor that respects game mode, I measured with my phone camera at 240 frames per second a difference I can't feel in play but can see on video, and it gives me a disproportionate satisfaction. I don't know whether this is a hobby or a symptom.
Controllers are another story. A USB controller polls at a frequency that depends on the device, and there are popular wireless controllers that add enough delay to ruin a demanding rhythm or platform game. For those who take this seriously there's SNAC, the connector that lets you plug an original controller into the board, a Mega Drive or Super Nintendo pad, without going through USB. It's the solution that makes sense and the one almost nobody uses, because it demands one more purchase, one more adapter, one more thing that has to be yours.
And the CRT. Right. The MiSTer can output RGB at 15 kHz, the signal old televisions expected, usually through a SCART cable from the analog board, with composite sync, and that opened a whole industry of people buying professional Sony monitors at prices that are an insult. I've seen someone pay more than fifteen hundred euros for a PVM to play Metal Slug. I found it ridiculous, and I still do. The CRT fetish has a part that's true, because the image from a cathode ray tube with real scanlines isn't the same as a shader filter on an LCD panel, and a part that's religion, because half of those who defend it never had a CRT around when they were twenty. Still, the first time I connected the MiSTer to an old television I had in the storage room, a 21-inch Philips with that typical flyback transformer hum, and saw Sonic with the picture all falling apart and alive at the same time, I shut up. I didn't write anything else that day.
But that's the hardware, and hardware is the easy part.
Software is where people lose their weekends. The SD card I mentioned is a start. Then there's the configuration file, with dozens of options with names that only make sense once you already know what you want. There are per-core configuration files, in the format MiSTer_corename.ini, that override the general values, and it took me a while to understand why my Amiga core behaved one way and the Mega Drive one another. The folder structure in /media/fat/games follows a convention you have to respect, or the menu finds nothing. And there's Update All, theypsilon's script that pulls the new versions of all the cores, all the scripts and several things you didn't know you needed from the repositories. The first run takes forever, downloads many gigabytes and leaves you with the feeling of having let a robot loose in a library. It's brilliant and dangerous. It's the reason most people manage to keep this updated, and also the reason that from time to time a new core breaks a game that was already working and you have no idea which of the thousand things changed.
There's SSH access too. The board has a network address, the user is root, and the factory password is "1". One. I didn't change it for about two months, which says everything about my security discipline on devices that aren't production servers. If you have the board on your home network with this default, change it. Now.
The legal part is where everyone does a bit of theater, and I think the theater is useful. The cores are open source, or at least available, and legal. The games aren't included. Asking for ROMs on the official channels is forbidden, and whoever asks gets a curt warning and a link to the policy. Everyone knows how it works in practice, and pretending otherwise is a bit ridiculous. But the rule has a purpose, which is keeping the project alive and out of the radar of people with lawyers, so I respect it even when it seems hypocritical. Whoever wants to play has to get their own dumps, from cartridges and discs they own, and that, to be honest, is more laborious than it sounds. The system BIOSes and disc images have to be in the right places, with the right names, and one swapped letter is enough to break it.
Some of the most impressive arcade cores come from Jotego, who works funded by supporters on Patreon, with early release for those who pay and opening to the public later. I've heard people outraged about this. I think it's fair. Someone who spends months rebuilding an arcade board, cycle by cycle, has a right to be paid for it, and the delay is a low price. What bothers me more is the economy around the project, the one that isn't his.
Because there is one, and it's ugly. There are resellers who assemble kits and charge double what the components would cost, with vague warranties and nonexistent support. There are third-party boards with dubious soldering that show up on Discord with the same problems and make the volunteers spend hours diagnosing what isn't a defect of the project. There are people who buy a ready-made kit, see the menu, and then ask where the games are. The polite answer is a link to the wiki. The tired answer is a sigh. Sometimes the community is harsh with beginners, and I think that has a cause that isn't malice: whoever has been answering the same question for six years has no patience left, and the wiki exists, and it's reasonably good. It doesn't excuse everything, but it explains.
And there are the effects nobody mentions in the videos. The first is time. I spent, conservatively, triple the time configuring versus playing in the first weeks, and I'm no exception. The MiSTer is a hobby disguised as a gaming hobby. The second is the collection. There's a disease of completing sets, of gathering every game from every system, that fills three-terabyte drives with things nobody will ever open, and that looks much more like hoarding than preservation. The third, and this one is subtler, is the effect on the market of what you want to preserve. Professional CRTs go up in price because the MiSTer created demand for them. Original cartridges, which some people buy to make dumps, go up too. The technology that was supposed to relieve pressure on old hardware, by letting you play without wearing out the original, ended up partly feeding it. I don't know if this is good or bad. I think it depends on who you are.
I'll contradict myself once more, because otherwise I'd be lying. I said that for most people the difference isn't noticeable, and I stand by it. But there was an afternoon with the Amiga core, with a Kickstart loaded from an image and a disk in HDF format that had taken me a while to find, when I heard the music from a game I hadn't thought about in decades, and it sounded the way it used to sound. Not like an emulator trying to sound like it. Like it used to. I don't know whether it was the filter, the timing or the wish for it to be so. Probably it was mostly the wish. But wishing, in a hobby like this, is also part of the hardware.
What this is remains unresolved for me. It isn't preservation in the museum sense, because nobody is archiving anything in a disciplined way. It isn't just nostalgia, because the engineering behind it is too serious. It isn't a business, despite the resellers. It's something made by people who discovered they like rebuilding a circuit more than using it, and who built around that a community that, on its best days, teaches and, on its worst, drives people away. I came in with the idea of playing Mega Drive with the right sound and stayed because of an argument at four in the morning about the timings of a sound chip, which I followed without understanding half of it and without wanting to leave.
As for the SD card, the lesson is the most boring one there is. Buy a decent card, from a known brand, and not some nameless thing found at an auction. Shut the board down from the menu, not from the plug. And keep a full image of the card on some drive, because I didn't, and that's why the Saturday was lost. I have nothing more to say about this, except that last week someone asked in the help channel whether it was worth waiting for the complete PlayStation core before buying the board. I didn't answer. I stayed looking at the question longer than I should have.