There are definitely still optimizations to be made but this is a lot faster than my previous experiment (plus they're better looking triangles in general)