Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 5 Sep 2020 21:16:40 +0200
From: magnum <>
Subject: Re: rar-opencl performance

On 2020-09-04 10:02, Solar Designer wrote:
> On Fri, Sep 04, 2020 at 01:50:02AM +0200, Axymeus wrote:
>> Alright, looks like Windows 10 is not properly reporting GPU usage
>> when running a PowerShell command... Looking at Afterburner I can see
>> that both cRARk and JTR are actually using 100% GPU, but not
>> constantly. There's a brief spike and a brief pause, but a much longer
>> pause between each spike with JTR. CPU usage is about the same. Here
>> is a visual comparison on the left of
>> the graph usage cycle of cRARk and on the right usage cycle of JTR.
> OK, so it's possibilities 1 or 2, but not 3.  (Of those I listed below.)
>> Unfortunately I won't be able to provide the archive being used. And
>> I'm not sure how to produce an equivalent, but I could give it a shot.
> magnum, do you have any advice for Axymeus on what parameters of the
> archive to look up and how, and how to replicate those?

Axymeus, please post the result of "unrar vt <archive name>".  My guess 
is this is a very large "stored" file.  I believe cRARk will process 
such files fully on GPU if possible, while we will not - as of now.  We 
could definitely implement that, I just never considered it worthwhile 
because it's so uncommon.


>> On Fri, Sep 4, 2020 at 1:31 AM Solar Designer <> wrote:
>>> On Fri, Sep 04, 2020 at 01:18:56AM +0200, Axymeus wrote:
>>>> cRARk goes up to 20K p/s on this archive.
>>> This can mean one of several things:
>>> 1. cRARk handles the archive smarter, rejecting candidate passwords
>>> earlier.  We do have early-reject in many cases, but apparently not
>>> working as effectively in yours.  If this is the case, then it's
>>> something magnum (a JtR developer) will want to improve in our
>>> rar-opencl format, I guess.  You might want to provide to him a copy of
>>> the archive - or preferably another "equivalent" archive with a known
>>> password and no sensitive content.
>>> 2. cRARk handles the archive incorrectly, rejecting candidate passwords
>>> early too aggressively, and might produce false negatives (miss the
>>> correct password).
>>> 3. cRARk implements full archive processing on GPU now - probably not
>>> the case.
>>> I guess #1 is most likely the case.
>>> What GPU utilization do you observe when cRARk is running on this
>>> archive?  And CPU too?  This might give us a hint.
>>> 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.