[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20120716094030.GB21271@openwall.com>
Date: Mon, 16 Jul 2012 13:40:30 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: xsha512-cuda & xsha512-opencl testing
myrice -
On Mon, Jul 16, 2012 at 04:55:45PM +0800, myrice wrote:
> This time I set password length > 9 and here are some results to fix the above:
"> 8" (9 or above) would have been enough. Not that this matters.
> Incremental xsha512-opencl 7970 password length > 8
> 1_1
> guesses: 0 time: 0:00:03:24 0.00% c/s: 14441K
Just a bit faster than current CPU code. Limited by transfers of
candidates from CPU to GPU, as expected.
> 100_100
> guesses: 0 time: 0:00:04:07 0.00% c/s: 48790K
OK.
> 10K_1
> guesses: 0 time: 0:00:04:00 0.00% c/s: 122552M
12M passwords/second, probably limited by transfers of hashes from GPU
to CPU, right?
> 1M_1
> guesses: 0 time: 0:00:03:58 0.00% c/s: 2788G
2.7M passwords/second, probably limited by large hash table lookups on
the host. Is this magnum-jumbo (without bitmaps) or bleeding (with
bitmaps)? I'd expect better speed with bitmaps. Either way, you need
to implement an equivalent of them on the GPU side. Not necessarily for
this hash type (where having so many hashes/salt is not typical), but
for saltless hashes mostly (where having millions of hashes and no salts
is common).
Thanks,
Alexander
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.
