MiSTer FPGA Forum 2026: 42,092 Peak, AI-Core Schism

There is a phpBB message board on the open internet that, for a few minutes in the summer of 2026, held more concurrent visitors than the population of a mid-sized town. At 1:20am on Friday, June 19, 2026, misterfpga.org logged 42,092 users online simultaneously — its all-time record, and a preposterous figure for a forum whose entire reason to exist is teaching people to program a sub-$250 development board into behaving like a Super Nintendo.

The MiSTer FPGA forum is not glamorous. It runs on the same open-source bulletin-board software that powered fan sites in 2005. Several of its boards will not let you post at all. Its front page throws a 403 at automated crawlers. And yet it remains the single most important coordination layer in field-programmable gate array retro gaming — the place where a new console core stops being a GitHub commit and becomes something you can actually load onto a microSD card and play. In 2026 it was also the place where the community nearly tore itself in half. This is the state of it.

The Forum in 2026: 42,092 Peak, 8,000-Plus Members

The numbers in the footer

phpBB prints its vital signs at the bottom of every board index, and MiSTer's have only moved one direction. A pinned thread still reads Over 5000 MiSTer Forum Members! — the board has comfortably doubled that. Live snapshots through August 2026 put registered membership north of 8,000, topics into the thousands, and the running post count deep into six figures. The exact numbers wobble depending on which server you catch: the stale mail. mirror lags the live index badly, and search-engine caches of the phpBB footer are notoriously unreliable. Treat the round figures as floors, not gospel.

Statistics --------------------------------------------------
Total members ....... 8,000+   (pinned: "Over 5000 MiSTer Forum Members!")
Total topics ........ 7,000+
Total posts ......... 94,000+
Most users ever online: 42,092  --  Fri Jun 19, 2026 1:20 am
-------------------------------------------------------------

42,092 online — the June 2026 spike

The one figure that is bulletproof, because the forum hard-codes it rather than recomputing it, is that record. And it deserves a raised eyebrow. Forty-two thousand humans do not sit down at 1:20 in the morning to simultaneously read about integer scaling and SDRAM timing. A spike of that shape, at that hour, is far more consistent with a scraper swarm than an audience — which is almost certainly why the administrators hardened the site against automated traffic around late May 2026 and started returning 403s to bots. The record is real; the implication that it represents live enthusiasm is the opposite of what the number actually documents. It is a monument to the AI-crawler era, sitting in the footer of a retro-gaming forum.

Locked boards and the signal-to-noise model

If you register today and try to reply in MiSTer Updates & Changelog, you cannot. That board — the one that matters most, because it is where core releases are announced — runs as locked developer megathreads. General Discussions was captured in a read-only posting state in October 2025, and the News-and-Announcements structure is explicitly project-updates-only. This is not neglect. It is a deliberate moderation model: keep the announcement layer clean and one-directional, and push conversation into the boards built for it, like the New Users Forum. On a hobby with this much overlapping misinformation, curating who can write where is the whole game.

What misterfpga.org Actually Is

A phpBB board, deliberately old-fashioned

The forum is a stock phpBB installation, chosen for exactly the reasons phpBB is usually mocked: it is boring, self-hosted, searchable, and it does not vanish when a company pivots. The board index splits into a handful of predictable zones. News & Announcements carries the changelog and the wiki pointer. A General cluster holds the New Users Forum, Development for MiSTer, and Guides & Videos. A Hardware cluster covers FPGA development boards, hardware setups, and a restricted Sellers board where only sanctioned vendors post. Below that sit the per-system Console Core Forums and Computer Core Forums, where a single game's glitch can get its own thread.

Announcements versus discussion

The split that gives the site its editorial value is the wall between announcement and discussion. When a developer ships a core, the changelog line goes into a locked thread; the arguing, bug reports, and "why does my CRT flicker" go somewhere you can reply. This is the opposite of a Discord, where a release notification is instantly buried under 400 messages of reactions. On the forum, six months later, the changelog line is still exactly where you left it, and still the top search hit.

MiSTer Organize and the DAT custodians

Underneath the visible boards runs a quieter maintenance economy. The MiSTer Organize project — coordinated through forum threads (topic 8287) and a plain misterorganize@gmail.com contact, with maintenance posts logged in March, July and August 2026 — keeps the DAT files, No-Intro naming, translation patches and RetroAchievements-compatible sets aligned with the cores. It is deeply unglamorous plumbing, and it is exactly the kind of work that only survives on a persistent forum, because it is a years-long conversation, not a moment.

From MiST to Sorgelig's Board

Till Harbaum, MiST, and analog-only accuracy

The lineage matters because it explains the forum's culture. Before MiSTer there was MiST, Till Harbaum's FPGA recreation of the Atari ST and Amiga. MiST was a purpose-built board with analog video only — VGA out, no HDMI, aimed at people who already owned CRTs and soldering irons. It proved the thesis that a field-programmable gate array could be the machine rather than emulate it, but it lived firmly in the tinkerer's basement.

June 2017 — the DE10-Nano bet

In June 2017, Alexey "Sorgelig" Melnikov introduced MiSTer on GitHub as a port of MiST to the Terasic DE10-Nano — an off-the-shelf Intel Cyclone V SoC FPGA board. The choice was strategic: the DE10-Nano gave HDMI output, an ARM companion CPU for the menu and file system, and, crucially, a board that Terasic manufactured at scale so nobody had to spin their own PCB. The forum grew up around that decision. Its default assumptions — HDMI is fine, the ARM side runs Linux, cores are shared as .rbf bitstreams over a script — are all downstream of 2017.

Taki Udon and the clone-board era

The board's second act was hardware pluralism. By 2024 the YouTuber Taki Udon's Retro Remake had shipped the MiSTer Pi, a low-cost Cyclone V clone of the DE10-Nano, and Heber Ltd had productised the community's Multisystem add-on into a finished console. The forum absorbed all of it without a schism — the Sellers board and the FPGA-development-boards board simply expanded to name Retro Remake, Heber and QMTech. That was the last time a major change reached the community without a fight. The next one would not go so smoothly.

The 2026 Core Explosion

3DO, Virtual Boy, Macintosh LC

2026 was the year the forum's changelog read like science fiction. srg320 (Sergiy Dvodnenko), already responsible for the SNES and Saturn cores, pushed a 3DO Interactive Multiplayer core into single-RAM beta — a machine long considered too hairy for FPGA recreation. The developer kitrinx shipped a Virtual Boy core in early August, complete with anaglyph and side-by-side 3D output for people who own the glasses or a 3D TV. The Macintosh LC core, tracked at github.com/MiSTer-devel/MacLC_MiSTer, got an optimisation pass around August 10 that fixed an HDMI PLL no-signal bug. Each of these landed on the forum first as a thread, then as a changelog line.

CPS3, RetroAchievements, DreamSTer

The arcade and infrastructure side moved just as fast. Capcom's CPS3 board — home of Street Fighter III — reached MiSTer around June. A RetroAchievements integration, built as Odelot's fork of the main firmware, brought hash-checked achievement tracking to roughly a dozen cores, with hardcore mode verified first on the NES and rolling out from there. And skmp, formerly of Reicast, dropped an experimental Dreamcast effort called DreamSTer that is barely a MiSTer core at all: a hybrid that hands the CPU to an ARM process and the PowerVR2 GPU to the FPGA, kills the MiSTer process to seize the board, and by its own author's admission "does not follow MiSTer conventions."

How a core reaches you

None of this is useful without delivery, and the delivery mechanism is the reason the changelog board is locked and sacred. A new core is a bitstream; the community's update_all script pulls it, along with arcade MRA sets, DAT files and now RetroAchievements hash databases, straight onto the microSD over SSH. The August 2026 roundup videos — one titled MiSTer News - Nintendo DS, Mac LC, Rainbow Islands, HiFi System & More (August 17) — are essentially annotated changelogs, each linking back to the forum thread that seeded it.

# A changelog line becomes a playable core, over SSH:
cd /media/fat/Scripts
./update_all.sh          # cores, arcade MRA sets, DATs, RetroAchievements DBs
# Update All 2.6 (theypsilon) adds Overscan: None / Low / Med / High / Max
# One "News & Announcements" post == one new .rbf on your microSD card

The AI-Slop Rift

What actually happened

Then somebody used a large language model to write cores. In mid-2026, developers began feeding MAME source code through tools like Claude Code and Codex to generate FPGA implementations of arcade games — Gaelco titles among them — at a speed no human hand-porting effort could match. To one camp this was the dam breaking: obscure boards preserved in hardware overnight. To another it was a category error being committed at scale, and it produced the ugliest thread the forum has hosted, bluntly titled Segregate Ai-Slop Cores to a Dark Corner of the Forum or Ban Them Altogether, which had run to four pages by August.

Jotego's indictment

The most credentialed objection came from Jose Tejada — Jotego, author of the JTCORES arcade collection — who did not mince words on X. His charge, in his own words, was that "All these are AI-generated ports of MAME using the JTCORES framework." His three-count argument, as relayed by the retro press, is that machine-translating an emulator into an FPGA turns the platform into "a second class MAME, that tries to emulate what the emulator does," drives skilled human developers away, and discourages the original hardware research — die shots, KiCAD schematics, hardware-level simulation — that made MiSTer accurate in the first place.

What is actually at stake

Strip away the tone and the fight is about the platform's founding promise. MiSTer's value proposition was never "more games" — that is what software emulation already does, cheaply. It was accuracy: an FPGA configured to match the original silicon so closely that input lag and edge-case bugs simply disappear. A core that merely recreates MAME's behaviour in gates inherits MAME's approximations while sacrificing the one thing FPGA was supposed to buy you. The critics are not luddites; they are defending the entire reason anyone spends €350 on a board instead of $50 on a Raspberry Pi. The rift is unresolved, and the forum's read-only levers are the only tools the moderators have to manage it.

What the Builders and Critics Say

Kotaku: why hardware simulation won

The clearest lay explanation of the stakes remains Mike Fahey's, writing for Kotaku. "No," he argued, "the MiSTer is the right-damn-now of retro gaming. By many metrics, FPGA hardware simulation surpassed software emulation a few years back." The mechanism, in his framing: "Instead of tricking a game ROM with software, you can program an FPGA to more or less be the original console hardware." The payoff is that the problems baked into software emulation, "like hard-to-avoid input lag, just go away." That sentence is the entire argument against AI-generated cores, written before AI-generated cores existed.

srg320 and the VRAM wall

The developers themselves are refreshingly candid about limits. On the 3DO core, srg320 told ReadOnlyMemo that "The main problem is VRAM emulation... Accurate VRAM emulation requires a certain amount of internal FPGA memory, which it doesn't have. Therefore, unfortunately, the core cannot be accurate on the MiSTer." He added that "the core itself is not complex and does not consume a lot of FPGA resources (it currently uses about 55% of the resources)" — the constraint is memory bandwidth, not logic. That is a hardware-first developer telling you exactly where the machine runs out of room, which is precisely the discourse the AI-core critics fear losing.

Lon.TV from the test bench

Lon Seidman's hands-on coverage bridges the two. On the 3DO betas he reported that "The 3DO core, developed by Srg320, is nearing completion and is currently available for testing on single RAM MiSTer devices," and that testing Road Rash "showed performance that appears consistent with the original hardware, though perhaps slightly less fluid than a stock console." That is what an honest FPGA frontier looks like: promising, playable, and still short of perfect — and documented thread by thread on the forum.

The Hardware the Forum Points At

The DE10-Nano and the Ultimate stack

The forum is agnostic about where you buy silicon, but the economics are worth spelling out. A bare Terasic DE10-Nano — Cyclone V SE, dual-core ARM Cortex-A9, roughly 110,000 logic elements — runs about $225, and by itself does nothing useful; you supply the SDRAM, IO board, case and power. The turnkey route is a vendor like Ultimate MiSTer, whose Ultimate MiSTer PRO bundles a genuine (non-clone) DE10-Nano with RAM and IO for €349.90 ex-VAT, with a Dual RAM variant at €379.90 for cores like 3DO and Saturn that demand the second memory module.

MultiSystem, RED, and ZERO

Above the kits sit the console-style complete builds. The MiSTer MultiSystem Kit lists at €309.90 ex-VAT — but note the trap the forum's Sellers board exists to prevent: that kit is a case, 128MB SDRAM, RTC, Wi-Fi/Bluetooth and an HDMI bridge, and it does not include the DE10-Nano. The finished RED and ZERO builds, which give you a NeoGeo-style box with everything inside, land around €599.90. For contrast, Heber's UK-built Multisystem 2 sells complete at £216 (digital) to £264 (analog) including VAT.

What you actually need to spend

The honest floor for a first MiSTer is the price of a DE10-Nano plus a RAM and IO board — call it the low-to-mid €300s once you add a case and a decent microSD. The clone boards have compressed this, and in a tightening 2026 silicon market they are increasingly the sane entry point rather than Terasic's original. If your instinct is that this is a lot of money to play Road Rash, the forum agrees with you and will happily point you at cheaper answers — which is the next section.

MiSTer vs Analogue vs Emulation

Analogue's closed FPGA consoles

MiSTer's nearest philosophical cousin is Analogue, which sells FPGA hardware as sealed, single-purpose consumer products rather than an open kit. Where MiSTer asks you to assemble a board and update it over SSH, Analogue ships a finished machine with firmware — the trade is convenience and polish for openness and breadth. Analogue's 2026 output has been substantial; its N64 machine, the subject of our look at how the Analogue 3D gained a free 2x upscaler, shows how far a locked-down FPGA product can be pushed. It will never run a Macintosh LC core the week a hobbyist writes one. That is the entire difference.

The software-emulation camp

The other pole is software emulation, which is cheaper, more portable, and — the MiSTer forum would insist — less accurate at the margins that matter. If your goal is the largest possible library on commodity hardware, the honest advice is a good frontend: our walkthrough of installing 200-plus RetroArch cores or a turnkey distro like the one in our Batocera setup guide will get you playing tonight for the price of a spare PC. The "you might as well play on a Raspberry Pi" jab in the AI-slop thread is not empty; for most systems below the 32-bit line, software is genuinely good enough.

The handheld flood

Then there is the handheld market, which has quietly eaten the low end of retro gaming while MiSTer argued about VRAM. A device like the one in our Miyoo Mini Plus breakdown plays everything up to the PS1 ceiling for pocket money, and premium Android units like those in our Retroid Pocket 6 versus Nova comparison push well past it. None of them match FPGA accuracy — but for a couple hundred dollars fewer, in your pocket, most players will never notice. MiSTer's answer has always been the same: accuracy is not for most players. It is for the ones who care about the margin. The forum is where those people live.

Forum, Discord, GitHub

The three-venue problem

The community's real structural story in 2026 is fragmentation across three venues that do not talk to each other cleanly. Development lives on GitHub, where the cores are versioned. Live help and chatter live on the official Discord, promoted from the forum's own X account and openly recommended by the project's documentation portal for "very active" real-time support. And the durable record — the searchable, years-deep archive — lives on the forum. The project's own history is now spread across all three, plus a docs site at mister-devel.github.io and a UK community hub, misterfpga.co.uk, last refreshed in March 2026.

Where the forum ranks

For all that, the forum's standing in the wider FPGA world is modest and worth stating plainly. Feedspot's 2026 Top 15 FPGA Forums list ranks the MiSTer FPGA Forum ninth — not the brief-sheet's oft-repeated "number two," which does not survive checking the actual page. Ninth is not an insult: it is the only consumer retro-gaming board on a list otherwise dominated by professional silicon communities like Intel's FPGA forum and Red Pitaya. A hobbyist board about pretending to be a Genesis, sitting mid-table among industrial engineering forums, is a strange and telling place to be.

Why the forum still matters

The obvious question is why the forum survives at all when Discord is livelier and GitHub is authoritative. The answer is permanence. A Discord message about a fixed HDMI PLL bug is gone in a day; the forum thread is still the top Google result a year later. The 2026 recap videos — including MiSTer FPGA: Everything That Happened in 2025, published January 4, 2026 — cite forum threads, not chat logs, because chat logs cannot be cited. The forum is the community's memory, and a community this dependent on hard-won accuracy cannot afford to forget.

What Happens Next: 6-12 Months

Governance and cores

Three things look likely by mid-2027. First, the AI-slop fight ends in policy, not a ban: expect a labeled subforum or a Sellers-style vetting tag for AI-assisted cores rather than outright prohibition, because the moderators already prefer partitioning to deletion, and their read-only levers are built for exactly this. Second, srg320's 3DO core graduates from single-RAM beta to a mainline changelog release — but with a permanent accuracy asterisk, because the VRAM constraint he described is architectural and does not go away with more effort.

Hardware and infrastructure

Third, the clone boards become the default entry point. With the silicon market tight and Terasic's original DE10-Nano perennially on backorder, the MiSTer Pi and MultiSystem kits will quietly become how most newcomers arrive, and the Sellers board will keep expanding to match. Fourth, RetroAchievements support consolidates: the Odelot fork either merges toward mainline or hardens into a semi-official companion build, with hardcore mode pushing past its initial dozen cores.

The forum itself

Fifth, and least happily, the forum's defensive crouch deepens. Expect more locked boards, more Discord-first support, and heavier bot mitigation as scrapers keep hammering the site — the same pressure that produced a 42,092 "peak" that was probably never 42,092 people. The most likely 2027 forum is a leaner one: an announcement-and-archive layer, deliberately quiet, with the live pulse moved to chat. That is not decline. For a project whose entire value is durable accuracy, a slower, better-guarded, permanent record is arguably the point. The phpBB board that outlived its era will outlive this one too.