instructions per clock have been going up, l3 cache sizes going up, core count going up, new wider instructions becoming available (avx 2 and 512 ), branch predictors improving.

overall speed increases are still nothing like the old days and spectre style mitigations have been eating away at the improvements