A 214-page teardown of how Commander Keen made a 1990 PC scroll like a console

The trick that made a PC scroll

Game Engine White Papers: Commander Keen is a 214-page, full-color technical book — free as a high-res PDF, with source on GitHub — dissecting how id Software's 1990 platformer pulled off smooth scrolling on hardware that wasn't built for it. It covers the 80286, EGA video, the engine architecture, and even the CGA version. The author frames it as personal: "It felt like returning to my early teenage years, debugging and experimenting with C and assembly code on a MS-DOS computer," a project that "took me over three years to complete."

The legendary core is John Carmack's adaptive tile refresh and EGA scrolling — redrawing only what changed instead of repainting the screen each frame. As one reader noted, Masters of Doom tells the human side of that breakthrough; this book supplies the engineering underneath it.

Why a slower console beat a faster PC

The most illuminating comment reframed why this was hard at all:

"It's not obvious to modern readers why a PC with significantly more powerful compute capabilities would struggle to keep up with significantly slower Nintendo hardware at the time for sprite rendering." — a reader on the hardware gap

Consoles like the SNES had dedicated sprite-and-scroll silicon; the PC had raw CPU and a frame buffer it had to wrestle by hand. Carmack's contribution was making general-purpose hardware imitate, in software, what Nintendo got for free in chips.

A different kind of cloning

The thread had its own meta-debate. Several readers noted the site's design looks unmistakably like Fabien Sanglard's famous game-engine black books, with one calling it "trampolining on his brand." Others widened the canon helpfully — pointing to Cosmodoc, a similar deep-dive into Cosmo's Cosmic Adventure, and to reconstructed Commander Keen source code — while a contrarian wondered aloud whether LLMs have made "all arcana about games engineering meaningless."

Preservation as a craft

The book is the opposite of the day's AI-slop submissions: three years of one person reconstructing exactly how a constraint got beaten, written for people who'll never ship for an 80286. Its value is precisely that it's slow, specific, and hand-made — a record of the moment software learned to out-think hardware it wasn't supposed to match. The LLM question in the thread answers itself: a model can summarize adaptive tile refresh, but it can't have spent three years caring about it.