It's amazing how many people didn't actually read all 4 parts of the article.
His argument has nothing to do with caching or prefetching, etc.
First, it's about random access. You can't prefetch a random fetch!
Second, he's measuring time, a perfectly valid thing to do. And the reality is when you lay your memory cells out in 2 dimensions it takes order of sqrt(n) time to fetch a random memory cell value, where n is the number of memory cells you're using.
Third, it turns out order of sqrt(n) time is the best you can do even if you had the best technology in the universe.
His argument has nothing to do with caching or prefetching, etc.
First, it's about random access. You can't prefetch a random fetch!
Second, he's measuring time, a perfectly valid thing to do. And the reality is when you lay your memory cells out in 2 dimensions it takes order of sqrt(n) time to fetch a random memory cell value, where n is the number of memory cells you're using.
Third, it turns out order of sqrt(n) time is the best you can do even if you had the best technology in the universe.