Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Thu, 23 Feb 2012 21:54:23 +0100
From: magnum <>
Subject: intel OpenCL (for CPU) "review"

I had a look at Intel's OpenCL SDK today. This one makes your Intel CPU
(or better, all CPUs and cores) work just like a GPU with OpenCL. For
some odd reason it does not support any of Intel's GPUs yet.

To my utter surprise, it installed without pain and worked like a champ
with the first OpenCL formats I tried. Performance is actually better
than our (thin dynamic) SSE2 version. I'm not sure how optimised the
latter is. Anyway this was on my three years old Core2 Duo laptop. I
really did not expect that.

$ ../run/john -test -fo:phpass-opencl
OpenCL Platforms: 1
OpenCL Platform: <<<Intel(R) OpenCL>>> 1 device(s), using device:
<<<Intel(R) Core(TM)2 Duo CPU     P8600  @ 2.40GHz>>>
Compilation log: Build started
Kernel <phpass> was successfully vectorized
Optimal Group work Size = 4
Raw:	17388 c/s real, 8735 c/s virtual

$ ../run/john -test -fo:phpass-md5
Benchmarking: PHPass MD5 [SSE2i 2x4x3]... DONE
Raw:	15936 c/s real, 15778 c/s virtual

However, it turned out I was lucky picking phpass - all other OpenCL
formats emits warnings like "Kernel <cryptmd5> was not vectorized"
during run-time compilation and they do not perform that well. Hopefully
this can be fixed. I really wonder why phpass is fine but not the
others. Anyway this is much more promising than I thought! And despite
bad performance, all formats do work fine, which means you can do some
testing or development on a machine lacking a hefty GPU card.

The SDK comes in RPM format from Intel but conversion to dpkg with alien
is trouble free. Here's an example:

BTW Samuele, I got the same LWS enumeration problem when trying this...


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.