[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 5 Mar 2016 20:47:59 +0100
From: Marek Wrzosek <marek.wrzosek@...il.com>
To: john-users@...ts.openwall.com
Subject: Re: devices for --dev= option
W dniu 05.03.2016 o 18:45, Matus UHLAR - fantomas pisze:
>
> On 05.03.16 17:57, Marek Wrzosek wrote:
>> I think it should be:
>>
>> john --list=opencl-devices
>>
>> to list OpenCL-capable devices.
>
> yes, but I somehow couldn't get --dev to complain without that one...
That's weird, I think it should always complain if value is not
numerical (or is referring to non-existent device) or not one of above
keywords.
>
>> And --device option is to choose one or more devices, if you have more
>> than one OpenCL-capable device, for actual cracking process. This option
>> allow only numerical values or keywords "all", "cpu", "gpu",
>> "accelerator" (or "acc" for short).
>
> yes, my question was if --dev only applies to opencl devices...
>
This --dev option was only mentioned in README-OPENCL. But I think that
this should work for CUDA too, if you have more than one nVidia with
CUDA support.
>> To use OpenCL you need to choose
>> format with OpenCL support.
>
> autoswitching formats to nvidia or opencl versions would be great...
>
I don't know which format JtR will use by default if there is GPU
equivalent but it would at least complain that there is faster format or
that there is another format for this type of hash. JtR is complaining
if OpenMP support for some hash is slower than using non-OpenMP version
with --fork, for example.
But if you list "hidden options" using command;
john --list=hidden-options
there is line that starts with "--format=CLASS". In my case this line
looks like this:
--format=CLASS valid classes: dynamic, cpu, omp
I can benchmark omp class of formats using this command:
john --test --format=omp
In your case there should be also some GPU-related classes. So, you
could use this class name instead of format name to give JtR a hint in
which class of formats it should look for the correct one, if there will
be more than one, then JtR will complain and I would need to use
--format option with the correct format name.
>> But I could be wrong, because I don't have
>> any OpenCL-capable device.
>
> I have nvidia capable of opencl and cuda... however openl seems to be
> faster. The CPU should be opencl-capable too, and as long as the card has
> about 225% performance of the CPU, using them all would be the best
> combination.
>
OpenCL is now faster than CUDA, that's why CUDA is disabled by default
in jumbo.
--
Marek Wrzosek
marek.wrzosek@...il.com
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.
