Try it on TurboWarp: https://turbowarp.org/1364682824?fps=60&stuck WASD and arrow keys, "O" for options The base texture is multiplied against the lightmap in order to light the scene. This project is expected to be slower, because the deferred rasterization process is demarcated at a fixed grid frequency on the base texture. This gives way for three assumptions and/or implications that the renderer can assume: The individual texels on the base texture are not further divisible and will always a contain a single color when rendered. Texture calculations, such as blending, exist on this texel frequency. The lightmap should not be of a higher frequency than the individual texels, otherwise lighting information can be lost or aliased. Although this increased frequency of color swaps on the rasterizer does tend to slow it down, it allows for the renderer to perform functions such as bilinear filtering at the rate of the texture resolution, as opposed to at a pixel resolution. This convenient decoupling will drastically increase the performance of these functions. In order to provide a small but certainly significant boost to the deferred swaps + blending + filtering is actually mipmapping, as it decreases the texel resolutions at higher distances, thus balancing the frequency of swaps + blending + filtering at various view distances. This also serves as a stronger optimization near full 480x360 resolution, and becomes less apparent at lower pixel resolutions.
