Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Sun, 10 Feb 2013 14:23:52 +0400
From: Solar Designer <>
Subject: dmg-opencl speed

magnum -

I just did this test:

First, I ran dmg2john on tests/openwall.dmg from Dhiru's VileFault
repository, with 156250 iterations.  I changed the iteration count in it
to 250000 (to match Jeremiah's as reported in his blog posting).  Then I
ran -i=all8 on it on bull's GTX 570 1600 MHz.  The speed is 486 c/s.
This is good, yet somehow it is significantly slower than what we're
getting for the PBKDF2-HMAC-SHA-1 in MSCash2.  Here it is 486*250000 =
121.5M iterations per second.  In MSCash2, it is ~30400*10240 = 311.3M
iterations per second.  OK, I see where a 2x difference comes from: we
need to derive a 256-bit AES key for the DMG, right?  But the difference
between 311.3M and 121.5M is 2.56x, not 2x.  Looks like there's another
~28% for us to squeeze by simply(?) making the code or maybe GWS more
similar to that of MSCash2.

The defaults were reported as:

Local worksize (LWS) 64, Global worksize (GWS) 9216

I am now testing LWS=256 GWS=81920 to match MSCash2's.  At these
settings (with the 250k iterations DMG), John is indeed very
unresponsive.  I wait for it to report status in response to my keypress
as I type this.  OK, it just did:

$ GWS=81920 LWS=256 ./john pw-dmg -i=all8 -form=dmg-opencl
Device 0: GeForce GTX 570 
Local worksize (LWS) 256, Global worksize (GWS) 81920
Loaded 1 password hash (Apple DMG PBKDF2-HMAC-SHA-1 3DES / AES [OpenCL])
guesses: 0  time: 0:00:02:17 0.00%  c/s: 595  trying: scharin1 - stachmar

Now this is 595*250000*2 = 297.5M, which gets close to MSCash2's 311.3M.


guesses: 0  time: 0:00:04:35 0.00%  c/s: 595  trying: stachmas - meddriel

Stable speed, but it takes 2 minutes to get a status update.


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.