Parallel and high-performance computing

Why are GPUs often effective for applying the same arithmetic operation to very large collections of independent elements?