mipmaps

Mipmaps are precomputed, downscaled versions of a texture used to improve rendering quality and performance in many 3D graphics systems. A mipmap chain consists of level 0, the original texture, and successively smaller levels where each level is typically half the width and height of the previous one, down to a 1x1 texel level. This arrangement provides a multi-resolution representation that helps the system sample textures at various distances and screen sizes.

During sampling, the GPU selects a mip level based on the texture’s footprint in screen space and the viewing parameters. Filtering is then applied within the chosen level, and, for smoother results, between adjacent mip levels (a process known as trilinear filtering). By using appropriately scaled textures, mipmaps reduce aliasing and shimmering that occur when a high-frequency texture is minified, while also improving cache coherence and reducing the number of texture fetches.

Generation and formats vary by platform. Mipmaps can be generated automatically at load time by graphics APIs (for example, OpenGL’s glGenerateMipmap or Direct3D equivalents) or supplied manually. Levels are typically created with a box or more refined prefilter, and most pipelines support both power-of-two and non-power-of-two textures, though NPOT support and mipmapping behavior can differ by hardware.

Usage considerations include memory overhead, since the full mipmap chain requires about 1.33 times the memory of the base texture. Mipmapping interacts with filtering modes and anisotropic filtering, which can further affect quality and performance. The concept was introduced in the 1980s as a solution to texture aliasing and sampling efficiency in real-time rendering.