Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 24 Mar 2012 23:21:03 -0400
Subject: Re: JtR: GPU for slow hashes

On Fri, Mar 23, 2012 at 9:37 PM, Solar Designer <> wrote:

> On Fri, Mar 23, 2012 at 10:27:09PM +0530, SAYANTAN DATTA wrote:
> > ... I want to apply for GSoC 2012 project idea "JtR: GPU for slow hashes
>  ".
> ...
> > Therefore ,based on my capabilities is it appropriate to apply for the
> > above project.Please reply.
> Please feel free to apply, but in order to have a chance to be accepted
> you need to demonstrate your ability to produce efficient code - not
> just faster than a straightforward implementation on CPU, but on par
> with what competing tools achieve on GPU (when a given hash type is also
> supported by other GPU-enabled tools) and/or many times faster than what
> a nearly optimal SIMD-enabled CPU implementation achieves.  You have to
> actually achieve that for at least one hash type before the student
> selection deadline (say, by April 15 or so - to give us at least a few
> days until the April 20 hard deadline to review and test your code).
> As an alternative or additionally to implementing support for a new
> hash type (for which we do not have any OpenCL or CUDA code currently),
> you may optimize some of the existing code - show that you're able to
> identify possible optimizations that the original author missed or did
> not have time to implement. ;-)
> There are also some hash types for which we have only CUDA or only
> OpenCL code - you may implement the other.  However, if you pick a
> "fast" hash, then you won't be able to show whether your code is
> efficient or not yet (that will require changes to JtR's formats
> interface first).  So either do it for a "slow" hash (as you had
> intended) or make it not your only task completed by April 15 (but
> rather an extra task for bonus points).
> You may obtain the current code with:
> git clone git://
> According to the table at we
> currently lack MSCash2 in OpenCL (but we have it in CUDA).  You may
> experiment with this one and compare your resulting speed numbers
> against hashcat's.  This is just a suggestion; please feel free to pick
> something else instead.
> Please let us know if you start work on any of this (and on what
> exactly), and please post weekly updates (like you would need to under
> GSoC, so it's good to start practicing that as well).
> Thanks,
> Alexander

Hi Alexander,

I have started working on OpenCL implementation of MSCash2 and I expect to
deliver the unoptimized version of codes on or before April 1(no April
fool).Also I would like to know the estimated time required for the
implementation above algorithm because its always better to test yourself
against the clocks.Since I am at the end of my semester I am having a
pretty busy schedule this week but I will try my best to deliver them as
early as possible.

Since I'm using AMD/ATI GPU I won't be able to test against the cuda
codes.Therefore, are there any standard results against which I can bench
my codes?  I am running on radeon hd 4890 .

If I face any problem I'll let you know.


Content of type "text/html" skipped

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.