|
Message-ID: <a695931d8354e3c1e744007eb0efc5ba@smtp.hushmail.com> Date: Wed, 16 Jan 2013 22:33:45 +0100 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: CUDA multigpu first draft for pwsafe On 16 Jan, 2013, at 22:22 , Lukas Odzioba <lukas.odzioba@...il.com> wrote: > I am posting this mostly as an archive. > Thanks to one of john users I got access to GTX 690. This is dual gpu > card and currently it uses only one of it, which is not good at all. > > I attached first draft of multigpu code for pwsafe format, we may not > commit that now, but it would be great to have it in the next official > jumbo. > For identical gpus code scales very well: > before: > Raw: 96376 c/s real, 97090 c/s virtual > after: > Raw: 192752 c/s real, 192752 c/s virtual > > I added "GPUS" macro to cuda_pwsafe.h, so we must set it before > compilation but, in the final version code will be able to use -dev > arguments which are now omitted. > Known problems: > -code will scale well only if gpus are identical > -it requires at least cuda 4.0 (this is not checked now) > -it wasn't heavily tested at all > > Good news is that almost all code changes were done in .cu file. > I'll try create proper patches for other formats. > > Lukas > <cuda_pwsafemultigpu.diff> Excellent, I will commit this to bleeding (sooner or later) and then we'll sort out any problems. magnum
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.