AmigaVision Fixes a 30-Year-Old Bug — and Nobody Noticed It Was Broken

Here is the shape of the problem. The Amiga's custom chipset — Agnus for DMA, Denise for display — was designed for NTSC televisions at 15.7 kHz horizontal scan. The pixels it generated were not square. They were never square. The Amiga 500's low-resolution mode produced a pixel-aspect ratio of roughly 1:1.2 — taller than wide. Every circle the hardware drew appeared round on a 4:3 CRT because the display's geometry stretched the signal horizontally by exactly the right amount. The television was part of the rendering pipeline. Take it away, and the math changes.

Every emulator since the mid-nineties has taken it away.

UAE, the workhorse Amiga emulator that Sorgelig's MiSTer Amiga core and virtually every software-side Amiga layer descends from, eventually inherited a set of default scaling assumptions. Those assumptions treated Amiga NTSC output as though the pixel-aspect ratio were 1:1, or close enough. On a 1080p flat panel, the result looks plausible — the proportions are almost right, the way a photo that's five degrees off-level looks almost straight until you put a ruler on the horizon line. Almost is the enemy. Almost means nobody files a bug.

"Almost right" is the most dangerous state in preservation. It is the state that never gets fixed because it never gets questioned.

The AmigaVision team questioned it. Their 2026.04.16 release, posted to the MiSTer FPGA forums in mid-April, includes a corrected NTSC scaling preset that adjusts the vertical integer scale and viewport crop to restore the pixel-aspect ratio the original hardware intended. The changelog mentions it as one line item among many — new game configs, updated WHDLoad entries, a handful of demo-scene compatibility fixes. No fanfare. No "we fixed a thing everyone got wrong for three decades." Just a corrected number in a config file.

The Schematic Reference: Denise Was Always Right

To understand why the fix works, you have to go back to what Denise actually does. The Amiga's display chip generates a bitplane-based framebuffer and clocks it out at one of two horizontal rates: low-res at roughly 7.09 MHz (NTSC) or high-res at roughly 14.18 MHz. The horizontal blanking interval, the sync pulses, the color burst — all of it conforms to NTSC RS-170A. The signal is legal. A 1985 Commodore 1084 monitor accepts it and displays it correctly, including the non-square pixel geometry, because the monitor's horizontal and vertical deflection circuits produce the right 4:3 raster.

A modern flat panel does not do this. It receives a digitized signal — via the MiSTer's HDMI output, via an emulator's render target — and maps it to a fixed pixel grid. The mapping requires knowing the source pixel-aspect ratio, and this is where the bug lived. The assumed PAR was wrong. Not catastrophically, not by a factor of two, but by enough to make Defender of the Crown's shields slightly oval, to make Speedball 2's arena subtly wider than the artists drew it, to make the demo scene's proudest pixel work read as though it were rendered on a machine with a different display geometry than the one the coders owned.

The AmigaVision fix sets the MiSTer core's vertical crop and integer scaling to values that reconstruct the correct PAR on a 1080p display. It is, mechanically, a handful of numbers in an INI file. Conceptually, it is an act of reverse-engineering the television.

Side-by-Side: Mechanism Mapping

The correction works at two layers. At the core level, Sorgelig's Minimig implementation on MiSTer faithfully reproduces Denise's output — the FPGA does not introduce its own scaling error, because it is clocking the same bitplane logic at the same rates. The error enters at the scaler level, where the core's raw output is mapped to HDMI. AmigaVision's preset intervenes here, adjusting the vscale_mode, vsync_adjust, and video_mode parameters so the scaler produces the geometry that a 1084 monitor would have produced.

Think of it this way: the FPGA is the Amiga. The scaler is the television. For thirty years, the television was slightly wrong, and nobody brought a ruler.

The demo scene noticed first. Posts in the "Best Amiga in 2026" thread on the MiSTer forums show users from groups like Desire and TTE running classic demos through the new preset and reporting that effects they assumed were asymmetric — copper-bar gradients, vector-ball routines — were actually symmetric all along. The artists had drawn them round. The display chain had squashed them. The forum discussion is technical and measured and reads like a group of people realizing that the version of history they carried in their heads was off by a few percent.

The artists drew circles. Thirty years of emulation showed them ellipses. The fix is not new code. It is a corrected assumption.

Why This Matters Beyond the Amiga

Every retro platform that originated on a CRT carries a version of this problem. The SNES outputs non-square pixels. The Genesis does. The PC Engine does. Each platform's community has, at various points, gone through the same correction cycle: someone measures the original signal, compares it to the emulated output, finds a discrepancy, patches the scaler. The Amiga case is notable for how long it took — and for how invisible the error was. Nobody complained because nobody remembered the correct proportions clearly enough to compare.

This is the central difficulty of display-fidelity preservation. The reference standard is a cathode-ray tube that no longer exists in most living rooms. The people who remember what the output looked like are working from decades-old sense-memory, and sense-memory is generous with forgiveness. You need measurement. You need someone to go back to the spec sheet, count the horizontal clocks, calculate the intended pixel geometry, and test the result against a calibrated original monitor. That is exactly what the AmigaVision team did.

It is unglamorous work. It produces a one-line changelog entry. It fixes a thing that most users will never consciously notice. And it is, in every way that matters, the actual work of preservation — not archiving the ROM, not emulating the CPU, but reproducing the experience as the original hardware delivered it to the original display. The ROM without the correct display chain is a manuscript without its typeface. It conveys the content. It loses the design.

Credit Where It Lands

The AmigaVision project is a community-maintained configuration and launcher suite for the MiSTer Amiga core. It does not get enough credit for being one of the most meticulous platform-preservation efforts running today. Every game entry includes per-title configuration. Every WHDLoad slave gets tested. Every scaling parameter gets measured against original hardware where possible. The team's forum presence is patient, technical, and generous with methodology. They show their work.

Sorgelig's Minimig core for MiSTer, which faithfully reproduces the Amiga's custom chipset in FPGA logic, provides the foundation that makes this kind of correction possible. If the core itself introduced scaling errors, no amount of INI tweaking would fix them. The core is clean. The scaler defaults were not. AmigaVision fixed the defaults.

Desire, TTE, and the broader Amiga demo scene deserve mention for being the first community to systematically test the new presets against known-good reference material. Demo coders know their pixel grids down to the copper cycle. When they say the proportions are finally right, that is a calibration report from people who built the test patterns.

A thirty-year-old bug, fixed by a handful of numbers. No patch notes in bold. No press release. Just a configuration file that finally tells the scaler what the television always knew: the pixels were never square, and now, at last, neither is the output.