Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 4 Dec 2020 16:34:00 +0100
From: Solar Designer <>
Subject: Re: dmg-opencl low performance/ low gpu utilisation

On Fri, Dec 04, 2020 at 03:34:12PM +0100, wrote:
> >Given the iteration counts and the speeds you report below, the above is
> >probably misreporting.  I suggest you install AMD's tools to monitor GPU
> >utilization and temperatures.  You should also be able to notice high
> >utilization through the noise and heat the GPU cards produce.
> Can you please specify the AMD tool you mention?

I'm not really familiar with this, but AMD has a series of Windows tools
to tweak GPU performance, which are overkill for mere monitoring but are
obviously usable for that as well.  Apparently, these are/were called
OverDrive, WattMan, and now Radeon Software Adrenalin 2020 Edition.  For
the latter two, see:

Quite possibly you already have a suitable tool installed along with the
AMD GPU driver you use.  You might want to look for an AMD Radeon
specific section in your Control Panel.

There are also generic (not limited to AMD GPUs) third-party freeware
tools like GPU-Z and HWiNFO64.

Since I don't use Windows and thus also not these tools, I can't
personally recommend a specific one of these, but I think any one of
them should work.

> >That's weird.  I don't recall us having different default verbosity
> >level on Windows.
> >
> >It's right in there in doc/OPTIONS (including in 1.9.0-jumbo-1):
> >
> >You can use "-v=5" for maximum verbosity.
> Sorry, saying "manual", but I was actually reading the
> no-parameter-output of john.exe

In 1.9.0-jumbo-1, that wasn't an exhaustive list of options.  One needed
to pass the cryptic "--list=hidden-options" to see the rest.  We've
recently made a change so that by default only "--help" is mentioned,
and when that is used then the full options list is shown.  This new
behavior should avoid that kind of confusion.

> Turns out that It needs at least v=4 (even explicit -v=3 does not show
> it). The output however is different to your example. It is a separate line.

Oh, yes, it does take at least "-v=4" to get these printed in
1.9.0-jumbo-1, and yes the formatting is different.

> The result of GPU2 is not stable, even though I interrupted the running
> attack. GPU1 seems rather stable:
> GPU2:
> Local worksize (LWS) 64, global worksize (GWS) 73728 (1152 blocks)
> ~521000 real, ~7800K virtual
> GPU2:
> Local worksize (LWS) 64, global worksize (GWS) 18432 (288 blocks)
> ~480000 real, ~ ~3300K virtual
> GPU2:
> Local worksize (LWS) 64, global worksize (GWS) 73728 (1152 blocks)
> ~522000 real, ~ 23500K virtual
> GPU1:
> Local worksize (LWS) 64, global worksize (GWS) 18432 (288 blocks)
> ~463000-466000 real, ~ 2600K -3000K virtual

Given these, I recommend that you specify explicit "-lws 64 -gws 73728"
on the command-line.


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.