Moreover, when some operation gets sped up by a orders of magnitude, it can be used for new things that you’d never consider when it was relatively more expensive.
Something that used to be precomputed offline can be done in real time. Something that used to work on a small samples can be applied to the whole data set. Something that used to be coarsely approximated can be computed precisely. Something that used to require large clusters of machines can be handled on customers’ client devices. Something that used to be only an end in itself can be used as a building block for a higher-level computation. Etc.
Something that used to be precomputed offline can be done in real time. Something that used to work on a small samples can be applied to the whole data set. Something that used to be coarsely approximated can be computed precisely. Something that used to require large clusters of machines can be handled on customers’ client devices. Something that used to be only an end in itself can be used as a building block for a higher-level computation. Etc.