Parallelism vs Sequential

A CPU is built for latency: a handful of heavyweight cores, each loaded with branch predictors, out-of-order engines, and huge caches — all to finish one complicated instruction stream as fast as physics allows. A modern flagship has tens of cores.

A GPU made the opposite bet. It packs over ten thousand tiny cores into the same silicon budget — each one slow and dumb, but all of them running at once. Raw throughput over per-thread cleverness.