Mode 7 was just a number.

Most people remember Mode 7 as a marketing term. Spinning floors. Fake 3D. The thing Nintendo put in every SNES promotional video and let the hardware box art imply was basically magic. It became a buzzword so thoroughly that the actual mechanism underneath it got lost.

Mode 7 was a background mode. That’s it. The SNES had eight background rendering modes, numbered zero through seven, each offering a different configuration of background layers with different color depths and capabilities. Mode 7 was the last one, and it traded away almost everything the other modes offered in exchange for one specific capability: affine transformation of a single background layer. Rotation. Scaling. Skewing. Applied to one tilemap, with no other background layers running simultaneously.

What affine transformation actually means in practice is that the PPU applies a matrix operation to the background before drawing it. Every pixel’s source coordinate in the tilemap gets transformed by a 2x2 matrix before being fetched, which lets you rotate and scale the entire background as a unit. By itself, with a fixed matrix applied uniformly across the whole screen, this produces a flat rotating plane. The spinning title screens, the world map in Super Mario World zooming out to show the full island, the rotating floors in F-Zero’s attract mode. It was impressive for 1990, but not actually three-dimensional.

The depth illusion came from a separate technique that the hardware supported but didn’t automate: HDMA, or Horizontal Direct Memory Access. HDMA lets a program update PPU registers mid-frame, synchronized to the scanline counter, without CPU intervention. Apply this to Mode 7’s transformation matrix and you can use a different matrix for every horizontal line on the screen. Scale the background more aggressively at the top of the screen than the bottom, and a flat tilemap suddenly appears to recede into the distance. The tiles near the horizon are scaled down to near-nothing, the tiles near the player are close to their actual size, and the graduated scaling in between reads as perspective. The track in Super Mario Kart isn’t geometry. It’s a tilemap being scaled one row at a time, with the scaling factor calculated to approximate a perspective projection.

This distinction matters for understanding what Mode 7 could and couldn’t do. The transformation applies to the background layer. Sprites, the separate layer used for characters, items, enemies, and anything else that needs to move independently, are completely unaffected by the Mode 7 matrix. The SNES has no hardware capability to rotate or scale sprites. If something on screen appears to be rotating or zooming, t’s either being faked through rapidly swapping pre-drawn animation frames or it’s the background pretending to be a sprite.

This is the constraint that explains some of Mode 7’s most distinctive staging choices. The giant Bowser fight at the end of Super Mario World, with the zooming, looming face bearing down on the screen, isn’t a sprite. It’s a Mode 7 background with almost nothing else on screen. Most of the screen is black in that fight, which is a direct consequence of having used the background layer for Bowser. You get one layer in Mode 7. If Bowser is the background, everything else has to be sprites, and the SNES sprite layer has its own limits on how much it can hold at once.

The stone golem boss in Super Castlevania IV works the same way. The boss appears to be a hulking creature that rotates and presents itself as a three-dimensional form. Under the hood it’s the background layer being rotated, with sprites composited on top for the parts that need to move independently. The presentation relies entirely on the player not looking too closely at which elements are actually rotating versus which are conventionally animated. Once you know what Mode 7 can and can’t do, you start reading these encounters differently.

Some games built their entire mechanical structure around what Mode 7 made possible. Super Mario Kart and F-Zero treated the transformed background as the world itself. The track is the tilemap, the HDMA-per-scanline scaling creates the driving perspective, and the whole game exists in the gap between what the hardware could technically do and what it could be made to appear to do. Pilotwings used Mode 7 for its landing zones and flight surfaces, combined with sprite scaling tricks for the aircraft. These games couldn’t exist on the NES and wouldn’t look the same on the PlayStation, because they’re built specifically for what this particular capability produced at this particular hardware budget.

Other games used Mode 7 as punctuation. World map transitions. Boss introductions designed to make something feel enormous for a few seconds before the actual fight began. These uses are almost purely theatrical, using Mode 7 for the feeling of scale and motion, then reverting back to normal sprites for the gameplay. The mode’s visual fingerprints, like sparse scenes, large flat surfaces, and big objects isolated against empty space, show up consistently across both approaches because they’re not stylistic choices, they’re constraints imposed by the hardware.

Mode 7 was narrow, powerful and had sharp edges. Developers who understood exactly where those edges were could make the SNES appear to do things it technically couldn’t. The name was marketing. The mode was math. And the math, applied carefully enough, could make a 16-bit console look like it understood perspective.