Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 19 Oct 2014 16:40:01 +0400
From: Solar Designer <>
To: Federico <>
Subject: Re: PKZIP and GPU acceleration

On Sat, Oct 18, 2014 at 04:19:29PM +0000, Federico wrote:
> Right now i'm trying to decrypt a pkzip hash on a c3.8xlarge
> This is a 1.7.9 jumbo with openmp:
> Benchmarking: PKZIP [32/64]... (32xOMP) DONE
> Many salts:     148070K c/s real, 4630K c/s virtual
> Only one salt:  34508K c/s real, 1078K c/s virtual
> At work:
> guesses: 0  time: 0:00:02:32 0.00% (3)  c/s: 28193K  trying: 29vgwbip - 
> 29v336s0
> How about this? Sounds good?
> On the official 1.8.0 i cant decrypt pkzip hashes (havent tryed a test 
> with fork=32)

You ought to use bleeding-jumbo, then --fork=32 will work:

It also has at least one bug relevant to pkzip archive cracking fixed,
compared to 1.7.9-jumbo-7.

When testing length 8 passwords, it gives about 80M c/s cumulative for
--fork=32 on 2xE5-2670.  For length 5, it's about 100M c/s cumulative,
and it completes in one and a half minutes (for printable US-ASCII).
This is for a "./configure --disable-openmp" build of JtR, testing
against  Its
trivial length 6 password normally gets cracked instantly, but I was
specifically locking to length 8 or 5 to avoid that.  BTW, 1.7.9-jumbo-7
fails to crack that password because of the bug I mentioned.


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.