Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 10 Jan 2013 04:05:06 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: des-opencl

On Thu, Jan 10, 2013 at 01:04:44AM +0530, Sayantan Datta wrote:
> On Thu, Jan 10, 2013 at 12:52 AM, Sayantan Datta <std2048@...il.com> wrote:
> > Using zero salt and 2501 iterations.
> 
> Edit: Even more is achieved by simply changing gws...
> 
> Benchmarking: Traditional DES [OpenCL]... DONE
> Raw:    2531K c/s real, 183500K c/s virtual

Cool.  Is it on bull or on your own 7970 (what Catalyst version if so)?

I'm afraid we'll need to keep the version without such unrolling (and
thus with key schedule array) as well, for other GPUs.  This is based on
Lukas' and my testing of the cracker posted in that reddit thread on
5850 (very slow) and GTX 570 (doesn't even complete running in sane time).
I think only some GPUs (only AMD GCN identified so far) tolerate
exceeding the instruction cache.

Alexander

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ