Nothing, except it's thousands (literally) of times slower than alternatives because it uses primitives instead of meshes and therefore doesn't lend itself well to GPU parallelization