Why Did Classic NES Games Fit in 128KB? Uncovering Retro Resource Secrets

I think the most amazing NES game is Super Mario, which has 32 distinct levels, each with hidden elements, yet its code section is only about ten kilobytes and the data section another ten kilobytes.

Even a simple "Hello World" binary today can be larger than that.

Although 128 KB may not seem tiny, the bulk of the storage is taken by resources rather than the program itself. Because the NES is an 8‑bit CISC machine, a single instruction can do a lot, so the program size is only a fraction of what a modern 64‑bit RISC binary would be.

Resources such as graphics, music, map data and level data dominate the size. An uncompressed NES sprite uses only a 2‑bit palette index, whereas modern true‑color images use 24‑ or 32‑bit depth, making the NES image data 12‑ to 16‑times smaller. Moreover, NES sprites are only a few dozen pixels wide, while today’s images can be thousands of pixels, resulting in a storage difference of five to seven orders of magnitude.

For audio, the NES uses 8‑bit MIDI‑style music, while modern games use PCM samples. PCM size depends on sample depth, rate, channels and length, whereas MIDI size depends only on the complexity of the score, giving another five to seven orders of magnitude saving.

Procedural‑generation 64‑KB 3D demos are a different story: they are small because they generate content mathematically rather than storing it, which requires little storage but high computational power.

Summary

Games heavily reuse tiles and palettes, with each pixel often occupying only 2 bits.

Developers meticulously optimise data structures to avoid any wasted bits.

Audio stores only modulation parameters for sound channels and reuses them whenever possible.

The code is written in assembly that directly manipulates hardware, leaving virtually no redundant instructions.