john-users - Re: CUDA the Ripper

Openwall GNU/*/Linux 3.0 - a small security-enhanced Linux distro for servers

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date: Mon, 2 Jun 2008 17:36:00 +0100
From: "Larry Bonner" <larry.bonner1@...il.com>
To: john-users@...ts.openwall.com
Subject: Re: CUDA the Ripper

> Why?  I disagree.  Most of the high-level code can stay intact.  In
> fact, for slow and/or salted hashes (when there are many salts), all of
> it can stay intact.
>
what i meant to say is, anyone who wants to load JTR onto the NVidia
cards is going to have to ADD alot of code using the CUDA api.

i was referring to Elcomsofts free md5 cracker in previous post, the
open source PoC is much slower than the closed source.

For example, PoC example, using 1 kernel

8600GT - ~30M k/s
8800GT - ~90M k/s

version 0.2, using 8 kernels

8600GT - ~64M k/s
8800GT - ~232M k/s

these results were all using "dumb" brute force.

using a dictionary or hybrid attack would degrade performance, since
you'd have to transfer more data between the host CPU and GPU.

-- 
To unsubscribe, e-mail john-users-unsubscribe@...ts.openwall.com and reply
to the automated confirmation request that will be sent to you.