Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 12 Nov 2014 19:29:30 -0900
From: Royce Williams <>
Subject: Re: Xeon Phi support?

On Wed, Nov 12, 2014 at 2:09 PM, Solar Designer <> wrote:
> On Tue, Nov 11, 2014 at 10:17:30PM -0900, Royce Williams wrote:
>> Given the recent sub-$200 pricing for the Xeon Phi 3151P,
> You mean 31S1P (that's S, not 5).

Yes, thanks.

> Note that you'll also need a capable motherboard (most won't run a
> Xeon Phi) and chassis supporting passively-cooled PCIe cards (there are
> inexpensive hacks around that, though).
> This is possibly the most affordable motherboard known to run Xeon Phi:

Thanks for the lead.  I had otherwise been looking at:

>> and the relative scarcity of inexpensive ZTEX boards :-)
> That's irrelevant.

In the context that I intended -- my own personal project to assemble
an affordable JtR platform for descrypt -- it's not irrelevant (unless
I'm missing your point).

>> what are the
>> potential and likely performance, and level of effort required to use
>> Intel Phi boards with John the Ripper? Just looking for some rough
>> approximations.
> For descrypt, up to 80M c/s in current benchmarks (with unreleased code).
> The level of effort depends on how much to support - which formats,
> which ways to access the Xeon Phi boards from host (OpenCL, OpenMP
> offload, OpenMP native).  OpenCL already sort of works, but is quite
> pointless to use currently, as magnum said.

Thanks for the overview - makes sense.

For the unreleased descrypt code, is that 80M c/s number dependent on
being fed candidate passwords from CPU, as is the case for
descrypt-opencl on 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.