Or if you’re scaling a large cluster of CPUs for parallel computations where a 0.02% increase can make a tangible difference in runtimes.