05-29-2010, 07:02 AM
this is an ATI related problem and comes up due the combination of two single problems. first, to get the full performance out of ATI it requires vectorizing. this means it is best crunching 4 passwords the same time and then check 4 results at the same time. this requires that the algo itself does not depend in any way of the input data (password). all operations can be done in the same order. second, the mysql algo is -not- based of a fixed step count as in md4, md5 or sha1. it depends on the password-length. so they are incompatible on the first sight. to get full performance this would require all 4 passes all have the same password-length. so if oclHashcat would be a brute-force only cracker this is no problem. but this is not the case since oclHashcat is designed to work with wordlists. the length on these words in the wordlist is arbitrary and so is the password-length.