[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABob6ir4pb6uONOM8HVabF_2AbSH+tpec_ut-QC_D0P_DKYUXw@mail.gmail.com>
Date: Sat, 11 Aug 2012 01:59:14 +0200
From: Lukas Odzioba <lukas.odzioba@...il.com>
To: john-dev@...ts.openwall.com
Subject: Re: Current -fixes GPU formats vs TS
2012/8/11 Solar Designer <solar@...nwall.com>:
> In trying to trigger the problem without running xsha512-cuda first, I
> added:
>
> cudaMemset(cuda_hash, -1, PWSAFE_OUT_SIZE);
>
> right after:
>
> cudaMalloc(&cuda_hash, PWSAFE_OUT_SIZE);
>
> This didn't make any difference in triggering the problem (nor in
> preventing it), but surprisingly it provided a 3% speedup (approx.
> 106k c/s to 109k c/s on bull's GTX 570).
>
> A memset with 0 also provides some speedup (a slightly smaller speedup?
> not sure).
>
> Any idea why? This might be a clue.
I wasn't able to reproduce this bug that way (running xsha first) but
it I was able to get 100% self test fails during ./john -test=0.
This speedup is weird what is c/s, 3% seems to be more than error
dispersion. I have no idea what is the source of this speedup. Maybe
some powersaving mechanisms inside cuda impelentation or device
itself.
On gtx470 I get 466->470 after adding memset.
Lukas
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.
