[Shootout-list] fannkuch

[Bengt Kleberg]

On 2005-05-21 06:53, John Skaller wrote:
...deleted
> I have just run some benchmarks on my AMD64 which are specifically
> run for about 5 seconds. On an otherwise unloaded machine, variations
> of 5% still occur in runs. 

i think this means that it is not neccessary to try and find runtimes 
that give us better than 5% variations. i still think it would be a good 
idea to find out what run times are needed to get a sought after accuracy.

my naive suggestion was to find the granulatrity of the timer and then 
make sure that runtimes are 10x that value. this i hope will give us 10% 
variation, and that would be 90% accuracy. i could be wrong.


bengt