For most of the 1990s, frequency was a fair shorthand. Two processors from the same family at 500 MHz and 800 MHz differed in almost nothing else, so the larger number reliably meant the faster machine. That relationship broke when designers ran into power density and could no longer raise frequency without raising heat faster than they could remove it.
What followed was a shift in where performance comes from: wider execution engines, deeper reorder windows, better branch prediction, larger and closer caches, more cores, and specialised blocks for video, encryption and matrix maths.
Instructions per cycle is the real lever
Instructions per cycle, or IPC, describes how much a core completes in each tick. A design with 30 percent higher IPC at the same frequency finishes a serial task roughly 30 percent sooner while drawing similar power. This is where most generational gains now originate, and it is why a modern laptop core at 3 GHz outruns a decade-old desktop core at 4 GHz.
Sustained power is the second lever
Peak boost frequency is a marketing figure that a chip may hold for seconds. What determines how a long export or compile feels is the sustained power the cooling system allows and how gracefully the chip degrades as it warms. Two identical processors in a thin chassis and a well-ventilated one can differ substantially on the same benchmark.
When comparing parts, look for benchmark runs that state the power limit and the duration of the test. A single-run score with no power context tells you very little.
What to compare instead
Use three figures. First, sustained multi-core throughput at a stated wattage. Second, single-thread performance, which still governs responsiveness in most desktop software. Third, performance per watt, which predicts battery life and thermal comfort more accurately than any frequency figure.