Ivan Prenosil wrote:

>
> The number 10000 is just myth.

It's a kind of empirical threshold. Of course, there's not precise number.

> The old algorithm simply means: the bigger cache, the slower.

With non-linear dependency, it should mean a problem :-)

> But this problem was discovered many years ago - with todays CPUs
> you will hardly notice any difference.
> (a year ago I tried to test whether the algorithm was improved
> in FB1.5 or FB2, but the difference between FB1 and FB2 was so small
> that it was not worth to remember the result)

When I was testing the new algorithm, the difference between v1.5 and
v2.0 was between 10% and 100% (depending on operation).


Dmitry