Hacker News Reader

This is all true, but I never encountered cache optimizations like that to cause more than a 2x difference on average and 5x in very rare extreme cases. This is very far from what you can get from algorithmic Big-O changes, where it is easy to get 1000x speedups.

View on HN

That's true. Having been briefly in academia, I only evaluated algorithms based on Big-O and just a bit of practical experience taught me that there tends to be a sweet spot, where the o(n^2.7) is both much faster then your naive implementation, but simpler to code, better constants, and better cache lines then the o(n^2.7003).

Reply on news.ycombinator.com