Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 14 Sep 2012 19:40:35 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: office-opencl

On 14 Sep, 2012, at 16:28 , Dhiru Kholia <dhiru.kholia@...il.com> wrote:
> Speed on AMD 6970,
> 
> ✗ ../run/john -fo:office-opencl -t
> OpenCL platform 0: AMD Accelerated Parallel Processing, 2 device(s).
> Using device 0: Cayman
> Optimal keys per crypt 65536
> (to avoid this test on next run, put "office_GWS = 65536" in
> john.conf, section [Options:OpenCL])
> Local worksize (LWS) 64, Global worksize (GWS) 65536
> Benchmarking: Office 2007 SHA-1 AES [OpenCL]... DONE
> Raw:	16591 c/s real, 595781 c/s virtual
> 
>> 150X speedup. Awesome!
> 
> It does take a while to complete (3 to 4 minutes) and X becomes
> unresponsive. Can this be fixed easily?

That's longer than some of the crappier cards I've tested with, I'm not sure why. There are several issues involved. The worst is the self-test scheme, this will hopefully be fixed in 1.8. Another time-waster is the auto-homing of GWS. This can be mitigated by following the advice and put office_GWS in john.conf, using a known good value (the one printed).

The unresponsiveness is another issue. This must to be fixed as I get ASIC hangs on the dreaded 7970 from too long kernel durations. I am planning to fix this in both RAR and Office but it's not one of the most inspiring tasks.

By the way, I made a vectorized version for AMD, expecting even better performance. The code works 100% fine on CPU (Intel or Nvidia driver), and on Nvidia GPU (although it hurts performance there). But the dang AMD GPU driver just segfaults for no reason at all. I'm near certain that is yet another driver bug.

magnum

Powered by blists - more mailing lists

Your e-mail address:

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