[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANJ2NMNsxp1hUqu0iUY4nu1FeJ8p3X9Y5J9Geh8WHV6H_CbmGQ@mail.gmail.com>
Date: Tue, 7 Aug 2012 23:46:02 +0800
From: myrice <qqlddg@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: Result of hard core password generation on 7970
Solar
On Mon, Aug 6, 2012 at 3:46 PM, myrice <qqlddg@...il.com> wrote:
> Solar
>
> On Fri, Aug 3, 2012 at 12:42 AM, Solar Designer <solar@...nwall.com> wrote:
>> Are you using the bitmap even with just 1 hash loaded? In the real
>> implementation, I guess you'll need to do direct comparisons when the
>> number of loaded hashes (for the current salt if applicable) is below a
>> threshold (like just 1 or 2).
>
> Here is the result of using local memory for hashes < 2048 and direct
> compare using register for hashes is 1 or 2:
> I set global work size to 2048*8 with all test cases:
>
> raw-md5-opencl with PG (Local memory)
>
> 1
> guesses: 0 time: 0:00:01:28 0.00% c/s: 826315K
>
> 1000
> guesses: 0 time: 0:00:01:45 0.00% c/s: 828407M
>
> 1000000
> guesses: 0 time: 0:00:00:52 0.00% c/s: 437215G
>
> The code is here:
> https://github.com/qqldd/myrice-JtR/blob/PG-test
>
> The file I changed:
> https://github.com/qqldd/myrice-JtR/blob/PG-test/src/opencl_rawmd5_fmt.c
> https://github.com/qqldd/myrice-JtR/blob/PG-test/src/opencl/md5_kernel.cl
Previous, in my code, the password generation have unnecessary read
global keys every iteration. I move these code out of two for
iterations. Also, I change global work size to 2048*32 which
contributes to better performance. Here is the result:
1
guesses: 0 time: 0:00:01:26 0.00% c/s: 1956M
1000
guesses: 0 time: 0:00:01:27 0.00% c/s: 1807G = 1807M
1M
guesses: 0 time: 0:00:01:25 0.00% c/s: 922618G = 922.6M
Thanks
myrice
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.
