Date: Tue, 21 May 2013 22:55:04 +0530 From: Sayantan Datta <std2048@...il.com> To: john-dev@...ts.openwall.com Subject: cmp_all on gpu for descrypt-opencl Hello, On Tue, May 21, 2013 at 5:36 PM, Solar Designer <contrib-apply@...nwall.com>wrote: > Sayantan - > > On Tue, May 21, 2013 at 01:50:35PM +0530, Sayantan Datta wrote: > > Fast hashes: > > > > 1. Experimenting with a dummy password generator for descrypt opencl just > > to get acquainted with new format. > > 2. Experimenting with cmp_all() on gpu for descrypt. > > You did not post about these experiments to john-dev yet, did you? > Perhaps you should. > > Thanks! > > Alexander > I have been able to run cmp_all() on GPU but the results aren't that good. In fact we have little performance penalty with cmp_all() on gpu. cmp_all() on cpu. Device 1: Tahiti (AMD Radeon HD 7900 Series) Optimal Global Work Size:524288 Benchmarking: descrypt-opencl, traditional crypt(3) [DES OpenCL]... DONE Many salts: 51615K c/s real, 246609K c/s virtual Only one salt: 28721K c/s real, 51609K c/s virtual cmp_all() on gpu: Device 1: Tahiti (AMD Radeon HD 7900 Series) Optimal Global Work Size:524288 Benchmarking: descrypt-opencl, traditional crypt(3) [DES OpenCL]... DONE Many salts: 48249K c/s real, 190241K c/s virtual Only one salt: 28230K c/s real, 50045K c/s virtual I think the overhead involved is too much and the kernel has lot of branching(that depends on user data). Also the amount of work is very less compared to overhead. One possible way to improve performance might be to load all hashes instead of just one per cmp_all() call. But that doesn't seems possible with current format. Regarding the password generation for descrypt, I haven't been able to get a fully working implementation yet. Thanks, Sayantan Content of type "text/html" skipped
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.