Compute-bound parallelism is not always data parallelism, for example a tree search algorithm would need to spawn/async on each tree nodes.