[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <e00493916d94c32cef0f6f90f412ccdf@smtp.hushmail.com>
Date: Thu, 06 Nov 2014 07:30:09 +0100
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: Discuss Mask-mode with GPUs and changes required for
its support.
On 2014-11-05 18:55, Frank Dittrich wrote:
> I think the GPU mask processor should support just a very limited
> functionality, i.e., just '?w' and ranges of characters, and core john
> should preprocess --mask and -1 ... -9 into a mask that the GPU formats
> understand.
> That way, --mask mode can be enhanced without the need to adjust all the
> GPU formats.
I'm sure that is Sayantan's plan. GPU will get preprocessed arrays of
characters per position, or something like that. Also, I believe CPU
will still process part of the mask. Eg. if your kernel wants ~250
million candidates per crypt_all() for a duration of about 100 ms, it
could load 256K words (produced by wordlist + CPU partial mask) and
apply the rest of the mask on GPU side, multiplying with yet another 1K.
In that example, transfer would be 1/1000th compared with CPU-only
generation.
I think we should normally aim for 100-200ms kernel duration, for many
reasons.
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.
