Date: Tue, 21 Jun 2011 22:47:38 +0200 From: Łukasz Odzioba <lukas.odzioba@...il.com> To: john-dev@...ts.openwall.com Subject: Lukas's Status Report - #6 of 15 This week: - I've tested newest nVidia drivers 270.41.19, and they are 2-3% slower than old 259 - I've tested new Cuda toolkit 4.0, it helped to solve some of "misterious errors" during kernel launches. - Found bug in md5-crypt cuda patch - it wasn't working on G80 architecture, because it has got separate instructions for reading from constant, and global memory. Compiler had a problem with it: Warning: Cannot tell what pointer points to, assuming global memory space And this time assumption was wrong which causes runtime errors. Fermi has unified memory operations/addresing so it worked fine. Bug is solved now. -Finished md5-crypt rev0. Password length is limited to 15, so I could use faster algorithm. Without any optimizations like loop unroling or use of shared memory i've achieved ~110k c/s. Next revision should easily double it but as always it's 2-3 slower than oclhashcat. -Created newer version of sha256-based crypt, it's not ready, need day maximum two to finish it. As md5 crypt i've limited password length. Next week: Finish sha512crypt. BTW https://mtgox.com/ has migrated to this format. Try to optimize previous patches.
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.