This project regards the recent discussion in the triangle filler community around the speeds of triangle filler tests being misaligned with actual performance in 3D engines. I am fairly sure this has to do with the proportions of the areas used in these tests. Here, the proportions of triangles of different areas in triangle speed tests is visualised. The red line represents aeᵇˣ which closely matches the area distribution. The domain of the graph is 0 to 86400 (86400 is the maximum size possible of a triangle with 3 verts within the scratch screen). The test uses the area of triangles chosen by picking a random points on the screen to use as vertices, as is standard in these tests. Clearly these tests favour smaller triangles, and as such are not a perfect indicator of true performance.
