Why games are so large: the anatomy of a footprint

Every gigabyte on a drive has a reason. Some reasons are unavoidable, some are a trade for speed, and a few are simply leftovers. Knowing which is which is what makes sensible reduction possible.

Textures and surfaces

Textures are the largest single category in most titles. A surface is described by an image, and a detailed surface needs a detailed image. When a scene contains a great many distinct surfaces, the collection of those images grows quickly. To make the result look sharp at different distances, a production often stores several versions of the same image at several resolutions. That technique gives smooth results, but it multiplies the count of files the drive must hold.

The same surface is frequently present in many places at once. A single wall, floor or prop can be tucked into a dozen different regions so that each region streams quickly without reaching across the whole archive. From the player’s point of view the duplicate is invisible. From the drive’s point of view it is a second and third copy of data that already exists.

Audio, language by language

Recorded dialogue is heavy. A campaign with a long script and a full cast can carry many hours of voice work, and each supported language adds its own complete set. Music and ambient beds add more. Because audio is played back continuously and cannot wait while it is decoded, it is often stored with little compression, which raises its footprint further.

Geometry, video and cache

Three-dimensional models and their supporting data make up another large block. Fine detail on a character or a vehicle is expensive, and a production keeps several levels so that distant objects can use lighter versions. Pre-rendered sequences, where the hardware is not asked to draw the scene live, are stored as video and can be enormous. Finally, modern titles build a shader cache: a store of compiled fragments that avoids repeated work. The cache is regenerated when drivers or the title change, which is why a fresh update can temporarily swell the figure on disk.

A player's hands resting on a backlit keyboard, the point at which a large asset library is finally used — Assets are stored once and used many times; the drive pays for every version a production decides to keep.

Trade-offs chosen on purpose

None of the above is a mistake. Duplicating art buys fast streaming. Keeping several detail levels buys a stable frame rate. Storing audio lightly compressed buys clean playback. Each decision makes the experience better in one dimension and heavier in another. A reduction strategy therefore has to be selective: it should target data that is either duplicated without benefit or never used at all.

What can be reduced safely

Three categories are usually fair targets. First, duplicated assets, which a careful storage manager can sometimes consolidate. Second, optional asset packs at higher resolution than the display can show. Third, language sets and bonus material a given reader will never open. Everything else belongs to the experience, and removing it would be neither wise nor honest.

The tools to address those categories already exist on the machine. The next sections describe the general principles of compression, then walk through the built-in facilities in Windows, macOS and Linux so that a reader can see exactly where each lever sits.