Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 26 Apr 2013 21:08:04 +0200
From: magnum <john.magnum@...hmail.com>
To: john-dev@...ts.openwall.com
Subject: Re: GPU support under Windows

On 26 Apr, 2013, at 18:00 , Dániel Bali <balijanosdaniel@...il.com> wrote:
>>  we're/I'm curious about performance of opencl formats on that
>> build. Could you do a comparison between speeds achieved on
>> linux-x86-64-opencl on Linux and your cygwin build?
>> Those with "-opencl" suffix in name should be enough.
>  
> My GPU is an AMD Radeon HD 7670M. And when I'm running a raw-md5-opencl format I get around 22.7K c/s.
> The GPU on bull@...nwall is afaik an AMD Radeon HD 7970. It does around 32K c/s.
> Both of these values are slowly but steadily increasing. Why is this?

When does it increase? During an --incremental session? That is normal and due to more overhead (in that cracking mode) in the start of a session.

> If this c/s value is a correct measure of performance then I would say the windows build doesn't perform too bad.

I hope you mean 32000K a.k.a 32M c/s, and anyway this is actually a very poor figure (a single CPU core should do 25% better than that on a 64-bit build and a good GPU should do over 100x that speed, ie. >3G c/s once we get candidate generation on GPU) but this is expected right now for raw "fast" formats like md5.

So to really benchmark stuff correctly right now, you should use a really slow format like wpapsk. My CPU can do 1374 c/s with that format (using one core with SSE2) and my laptop GPU (nvidia GT 650M) can do 6-7K c/s. The 7970 achieves over 130K c/s.

Anyways, congrats and kudos for getting it to build in Cygwin! I'll look into adding your make target in the git tree.

magnum

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.