|
Message-ID: <CABob6ippwaL-kDTU7GJ0w6jQMta2oYpFYK_YCNoKB0r65QPtnA@mail.gmail.com> Date: Thu, 4 Aug 2011 06:06:45 +0200 From: Lukas Odzioba <lukas.odzioba@...il.com> To: john-users@...ts.openwall.com Subject: JtR CUDA patch Hi all! I've prepared patch adding nVidia GPU's support for the following formats: phpass, cryptmd5, cryptsha256, cryptsha512 Performance is rather poor, here's a few benchmarks on my GTX460: phpass 292k c/s cryptmd5 391k c/s cryptsha256 7800 c/s cryptsha512 5600 c/s For now it doesn't utilize multigpu's, but it will be added in near future. To get best performance hand made tweaks are needed: For Fermi's change -arch=sm_10 to -arch=sm_20 in Makefile Each CUDA format needs specific THREADS and BLOCKS constants defined in *.h files. There have got huge impact on performance so you should experiment with them, here's important heuristics: -BLOCKS should be multiple of device SM (Streaming Multiprocesors) -THREADS should be multiple of 32 (usualy more than 128 is needed, up to 512 on Fermi's). Please remember to rebuild after changes. Patch is uploaded to wiki (http://openwall.info/wiki/john/patches) as john-1.7.8-allcuda-0.diff GPU patches have got page on wiki too: (http://openwall.info/wiki/john/GPU) If you having any problems with this patch feel free to contact me. Any kind of feedback is welcomed. Lukas
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.