(05-26-2012, 10:19 PM)KT819GM Wrote: On dict or rule based attack you have to pull data to gpu, so you have plenty of it's power to utilize. On brute force atom already made optimizations to use all of gpu to process as much characters as possible. On the fly filtering will slow down calculation as same like any other process witch uses gpgpu. Somehow should be done with template, like on other thread for sl3 you have always same dataset so can make some template for not to check, but really would like to see how much combinations would be 'saved' if skipping 10+ same characters in a row
I understand what you are getting at ... I think !
It may be that atom could optimise the brute force, which is what I am hoping for.
I did a very basic and simplistic calculation to try to work out the gain and I "think" if you had to search the entire keyspace using an optimised version it "should" possibly save you 18%.
I could be completely wrong though and it is based on 8 upper alpha and filtered to remove all but unique characters consecutively.