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

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