Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 22 Sep 2012 01:44:59 +0530
From: Sayantan Datta <std2048@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: bitslice DES on GPU

On Sat, Sep 22, 2012 at 1:41 AM, Sayantan Datta <std2048@...il.com> wrote:

>
>
> On Sat, Sep 22, 2012 at 12:51 AM, Solar Designer <solar@...nwall.com>wrote:
>
>> Hi Sayantan,
>>
>> On Sat, Sep 22, 2012 at 12:31:26AM +0530, Sayantan Datta wrote:
>> > I was experimenting with DES_BS_EXPAND set to 0 and using different
>> types
>> > of memory .
>> >
>> > 1. B[] in local memory and K[] in private register space:
>> > With this combination I'm getting speeds of around 24M c/s on 7970 and
>> 5M
>> > c/s on 570.
>> >
>> > 2. B[] in private register space and K[] in local memory
>> > I'm getting 12M c/s on 570 while I couldn't get it working on 7970.
>> Getting
>> > hash fails every time. Most likely the problem is with referencing B[]
>> > array in z(p) macro.
>>
>> That's curious.  Even more interesting would be speed numbers with the
>> overhead mostly excluded - that is, use this test vector:
>>
>>         {"..X8NBuQ4l6uQ", ""},
>>
>> set the iteration count e.g. to 2501 (any odd value should do), and
>> multiply the reported c/s rate by 100 (if you picked 2501) to get the
>> descrypt equivalent cracking speed.
>>
>> As to the overhead, we'll need to deal with it by other means later.
>>
>> I just did some math, and I think that your 24M with overhead may
>> correspond to around 65M in the without-overhead test.  Please confirm
>> or disprove. ;-)
>>
>> 1/(1/39+1/41) = 19.99
>>
>> 1/(1/39+1/60) = 23.64
>> 1/(1/39+1/65) = 24.38
>>
>> 39M is my guess as to the "overhead speed" alone (without crypto), based
>> on the 19.9M "with overhead" speed you reported and my 41M
>> without-overhead test.  Assuming that this "with overhead" speed
>> remained the same, it'd take around 65M without-overhead speed to
>> reach/exceed 24M reported for overhead+crypto.
>>
>> Thanks,
>>
>> Alexander
>>
>
> Hi Alexander,
>
> In the previous test Global no. of work items were half of this time. So
> the overhead is double in this test than the last one.
> Setting 2501 I'm getting 35M without any overhead. So your guess of 65M is
> accurate.
>
> here's the math:
>
> 1/(1/(39x2) +1/65) =35.4
>
> Regards,
> Sayantan
>

Sorry my math is wrong. Result is weired.

Regards,
Sayantan

[ CONTENT OF TYPE text/html SKIPPED ]

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ