[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20120913203230.GA1187@openwall.com>
Date: Fri, 14 Sep 2012 00:32:30 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: bitslice DES on GPU
Hi Sayantan,
On Mon, Sep 10, 2012 at 01:12:54PM +0530, Sayantan Datta wrote:
> Here's the result of keeping B in local memory.
[...]
> Benchmarking: DES BS [128/128 BS XOP-16]... DONE
> Many salts: 19942K c/s real, 130023K c/s virtual
> Only one salt: 15617K c/s real, 45875K c/s virtual
This is a nice improvement compared to your initial try. It's CPU-like
speed now (comparing CPU vs. GPU per-chip), except that the program is
less interactive than it is when using CPU.
To see the speed of the hashing loop on GPU alone, I've tried running
this with 2501 iterations (and my special test hash). I got:
Raw: 410401 c/s real, 209715K c/s virtual
(no single-salt vs. multi-salt since I only had one test hash, and thus
one test salt).
So that's about 41M c/s at 25 iterations if you somehow manage to remove
the overhead. In other words, the overhead still corresponds to about
50% of total running time.
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.
