Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20130627123353.GC20753@openwall.com>
Date: Thu, 27 Jun 2013 16:33:53 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: latest version of descrypt project on reddit

Sayantan,

> On Tuesday 25 June 2013 09:21 AM, Sayantan Datta wrote:
> >Please post the link to the latest version of the descrypt project on 
> >reddit that you mentioned during our last irc chat.

It's not public, and it's not ready to be made public because that would
misattribute the various hacks to the original author (the comments would
need to be edited first).  Instead, I've placed it in your home dir on
bull - please take a look.

On Wed, Jun 26, 2013 at 07:13:06PM +0530, Sayantan Datta wrote:
> I ran the descrypt project on my 7970. I found that it was doing only a 
> single iteration inside the kernel.
> http://www.reddit.com/r/crypto/comments/162ufx/research_project_opencl_bitslice_des_bruteforce/
> 
> For single iteration its speed was:
> 
> Run time: 0.114362 s
> Rate: 4694.471722 Mkeys/s
> Time to search keyspace: 177.655758 days

That's better than what I saw prior to making any changes at the time.
Apparently, new Catalyst optimizes the original code a lot better now.

Anyhow, with my somewhat optimized code new Catalyst is actually slower
than whatever version we had on bull at the time.  I got 6800+ Mkeys/s
at the time, now only getting 6500+ Mkeys/s.  The original kernel gives
me 5700+ Mkeys/s now.

Optimized:

Run time: 1.316827 s
Rate: 6523.205328 Mkeys/s
Time to search keyspace: 127.851246 days

Original:

Run time: 1.501679 s
Rate: 5720.220933 Mkeys/s
Time to search keyspace: 145.798552 days

Hmm, I don't know why your "Run time" is 10x lower.  Maybe there was
some other change.  (I started hacking based on Lukas' port to Linux.)

> So I modified the code a little bit to put it under 25 iteration. 
> Surprisingly the speed got reduced by 1/50 times instead of 1/25!! I'm 
> trying to figure out why?

Yes, please do try to figure this out.

Thanks,

Alexander

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.