Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 13 Sep 2011 19:35:40 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: Speed of jtr on your machine?

Jim, Bugtrace -

On Tue, Sep 13, 2011 at 10:16:48AM -0500, jfoug wrote:
> Shouldn't linux-x86-64 build with sse-intrinsic code?

Yes, but with gcc 4.4+ only.  I guess Bugtrace's older version of Ubuntu
has a pre-4.4 version of gcc.

That said, my build of 1.7.8-jumbo-5 for linux-x86-64 with gcc 4.5.0
also prints "Raw MD5 [gen]", on E5420 2.5 GHz:

Benchmarking: Raw MD5 [gen]... Using raw-md5 mode, by linking to md5_gen(0) functions DONE
Raw:    13203K c/s real, 13335K c/s virtual

Benchmarking: PHPass MD5 [phpass-md5]... Using phpass mode, by linking to md5_gen(17) functions DONE
Raw:    11376 c/s real, 11376 c/s virtual

With icc (well, using Simon's sse-intrinsics.S file in my case since I
don't actually have icc installed):

Benchmarking: Raw MD5 [gen]... Using raw-md5 mode, by linking to md5_gen(0) functions DONE
Raw:    16065K c/s real, 16225K c/s virtual

Benchmarking: PHPass MD5 [phpass-md5]... Using phpass mode, by linking to md5_gen(17) functions DONE
Raw:    15864 c/s real, 15864 c/s virtual

> >From: Bugtrace [mailto:bugtrace@...il.com]
> >
> > # ./john | grep 'ver:'
> >John the Ripper password cracker, ver: 1.7.8-jumbo-5 [linux-x86-64]
> > # ./john --format=raw-md5 --test
> >Benchmarking: Raw MD5 [gen]... Using raw-md5 mode, by linking to
> >md5_gen(0) functions DONE
> >Raw:	5898K c/s real, 5957K c/s virtual
> >
> > # ./john --format=phpass-md5 --test
> >Benchmarking: PHPass MD5 [phpass-md5]... Using phpass mode, by linking
> >to md5_gen(17) functions DONE
> >Raw:	3552 c/s real, 3552 c/s virtual

Bugtrace - for better speed at these specific hash types, you either
need a newer version of gcc (at least 4.4) or you should be building
using the linux-x86-sse2 target (requires 32-bit compatibility libraries
to be installed on your system).  The former (gcc upgrade) should result
in better performance.  Well, or you may use icc instead of gcc, and use
the linux-x86-64-icc make target then.

> >It looks like not as fast as jtr 1.7.6_jumbo_12 x86 with sse2 .
> >
> > $ ./john | grep version
> >John the Ripper password cracker, version 1.7.6-jumbo-12
> > $ ./john --format=raw-md5 --test
> >Benchmarking: Raw MD5 SSE2 [raw-md5 SSE2 16x4]... DONE
> >Raw:	9668K c/s real, 9668K c/s virtual
> > $ ./john --format=phpass-md5 --test
> >Benchmarking: PHPass MD5 SSE2 [phpass-MD5 SSE2]... Using phpass mode,
> >by linking to md5_gen(17) functions DONE
> >Many salts:	6144 c/s real, 6144 c/s virtual
> >Only one salt:	6083 c/s real, 6144 c/s virtual

It's primarily the make target difference, not JtR version difference.

For other hash types (not based on MD5), things may be very different.
For example, your build of 1.7.8-jumbo-5 is likely significantly faster
than your build of 1.7.6-jumbo-12 at DES-based hashes.

Alexander

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.