Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 23 Apr 2014 02:29:02 +0200
From: magnum <>
Subject: Re: recommended way to crack a descrypt hash in a multi-gpu
 setup (HD 7990)

On 2014-04-23 01:30, yungai wrote:
> in recent thread I read about the --fork feature in conjunction with
> GPUs but simply trying --fork=2 didn't work for me (running a current
> bleeding-jumbo git version).
> So I wanted to ask you what the recommended way to crack a hash with
> multiple GPUs currently is?

You need to supply a list of devices. Use --list=opencl-devices to see 
the device numbers:

$ ../run/john --list=opencl-devices
Platform #2 name: NVIDIA CUDA
Platform version: OpenCL 1.1 CUDA 4.2.1
	Device #0 (5) name:	GeForce GTX TITAN

The numbers you need are the ones between parens, in the above example 
it's 5. You probably have 0 and 1. So you'd do this:

./john -fork=2 -dev=0,1 -format=descrypt-opencl ...

or just use the shortcut for "all GPUs":

./john -fork=2 -dev=GPU -format=descrypt-opencl ...

> What kind of performance can one expect when cracking descrypt on a HD
> 7990? I remember seeing benchmarks on john-dev(?) with ~102000 Kc/s but
> I can't seem to find that benchmark anymore

Performance has varied a lot over time with different driver versions. 
For a long period, AMD driver only got slower but I think the situation 
is better now.

> I get on my HD 7990 (with device 0):
> ~47000 Kc/s

That would be the "many salts" benchmark figure. Attacking just one salt 
will be more like 36000Kc/s.

> so that would be near the value I remembered if you multiply that by two
> for the second GPU.

Yes. You might be able to push it a little with some LWS/GWS tuning, 
sacrificing responsiveness. See doc/README-OPENCL


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.