[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 18 Apr 2012 14:12:40 +0200
From: Samuele Giovanni Tonon <samu@...uxasylum.net>
To: john-dev@...ts.openwall.com
Subject: Re: System Responsiveness and GPU temps During Cracking
on GPU
On 04/18/12 13:32, SAYANTAN DATTA wrote:
> Hi Samuele,magnum and others
>
> I was stress testing my gpu and MSCASH2 algorithm for some real
> scenarios. So I loaded a 6 digit password("hd4890") hash in the file
> pw. This is what I loaded.
>
> sayantan:64d3e78d25d36de331cf680c6603f89b
>
> Then I ran:
>
> ./john pw -fo=mscash2-opencl
>
> After running for about 20 minutes i.e. nearly 350 crypt_all() calls on
> my gpu it started producing errors. The clEnqueNDRangeKernel() started
> failing. My GPU was running really hot(max 88C) and I think this is the
> reason for the failure. Also the desktop becomes unusable for 20
> minutes which is certainly very annoying.
>
> Hence something must be done to resolve these issues.
>
> Limiting the number of work groups helps in increasing system
> responsiveness and reduce temps but the performance drops drastically.
> Are there any ways to limit the number of compute units. ie if I have 10
> CUs then only 9 CUs or less would be used for computation.
hello Sayantan,
i think that what you are seeing is not related to a bad behaviour of
the code for that format but rather an hardware problem .
I'm assuming your system is configured with not so many customization
and if you do some gaming it will response good (even if a bit slow).
So unless you have overclocked your gpu, you should not get errors and
unresponsiveness by just having "not optimized" code: you could
experience some slowness on the desktop, true, but linux should scale
to not hang.
So i'm guessing there's a hw problem and the only way to reduce your
temp is to try to clean the gpu from the dust, putting your desktop near
circulating air, keeping away from hot places and if in doubt, find a
"$2 fan" to make air circulating inside your desktop, even a bit of air
might low your temp by 5-10 degrees.
However this is not really jtr related but more likely a hardware
problem that need to be solved .
On the other hand by limiting number of work group you are limiting your
temp because your making less work per unit in your GPU; of course
performance will drop.
If you find a way to - on the same hardware = increase performance while
limiting gpu/cpu power and therefore temp, then you would have something
to be billionaire :-)
Cheers
Samuele
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.
