Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 14 Oct 2012 19:50:47 +0400
From: Solar Designer <>
To: "[Jano]" <>
Subject: Re: Performance with SSE2 support with WPA-PSK and openCL error

Hi Jano,

Why are you posting this to john-dev?  You appear to raise typical
end-user concerns, so your message would be on topic for john-users, not
for john-dev.  The latter is for discussions around JtR source code.
Now that the message is already on john-dev, let's continue this thread
in here, but going forward please note that this kind of stuff is more
suitable for john-users.  Thanks.

On Sun, Oct 14, 2012 at 02:05:14PM +0200, [Jano] wrote:
> Benchmarking: WPA-PSK PBKDF2-HMAC-SHA-1 [32/64]... DONE
> Raw:	352 c/s real, 352 c/s virtual

You have a non-OpenMP build here, hence you're using just one CPU
core here.  You may improve speed (to around 2000 c/s maybe, depending
on your CPU type) by building with OpenMP (see README).

We have no SSE2 support for WPA-PSK yet, this is true. :-(

> OpenCL error (CL_OUT_OF_RESOURCES) in file (opencl_rawmd4_fmt.c) at line (300) - (failed in clEnqueueWriteBuffer buffer_keys)
> OpenCL error (CL_INVALID_CONTEXT) in file (rar_fmt.c) at line (367) - (Error Unmapping aes_key)

You ran plain --test, which is known to fail in GPU-enabled builds - see
doc/BUGS in 1.7.9-jumbo-7.  The GPU-enabled WPA-PSK format was not yet
benchmarked in your test run (the program crashed sooner than that
format would have been benchmarked).  You probably want to run:

john --test --format=wpapsk-opencl

You will also want to make an OpenMP-enabled build (both OpenMP and
OpenCL at once), since the wpapsk-opencl format benefits from that as well.
This should give a c/s rate in tens of thousands on a reasonable GPU.


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.