cuda performance lower then with cpu-only (from svn)

[GoogleCodeExporter]

What steps will reproduce the problem?
1. build and install pyrit
2. build and install cpyrit_cuda
3. cd ~; pyrit benchmark

What is the expected output? What do you see instead?

I expected this 'Nvidia Quadro FX 360M' to deliver more then the SSE2 dual-core 
(intel T7500)

What version of the product are you using? On what operating system?

SVN version on Debian Wheezy as package from repo has no cuda/opencl supplied.

Please provide any additional information below.

None avaailable at this time.

Original issue reported on code.google.com by joris.la...@gmail.com on 8 Sep 2012 at 9:37

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

Seems similar to issue 122

Original comment by joris.la...@gmail.com on 12 Sep 2012 at 6:32

[GoogleCodeExporter]

Tried by setting list_ncpus = '1' in cpyrit/config.py since this is a dual-core 
machine

**** results are below ****

pyrit benchmark
Pyrit 0.4.1-dev (svn r308) (C) 2008-2011 Lukas Lueg http://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3+

Calibrating... 
Running benchmark (831.1 PMKs/s)... - 

Computed 831.07 PMKs/s total.
#1: 'CUDA-Device #1 'Quadro FX 360M'': 509.9 PMKs/s (RTT 3.0)
#2: 'CPU-Core (SSE2)': 410.4 PMKs/s (RTT 2.9)

Original comment by joris.la...@gmail.com on 12 Sep 2012 at 5:53

[GoogleCodeExporter]

Just verified, hyper-threading is disabled. Validated cores vs siblings on 
output of /proc/cpuinfo to be equal. Used 
http://dag.wieers.com/blog/is-hyper-threading-enabled-on-a-linux-system as 
reference.

Original comment by joris.la...@gmail.com on 12 Sep 2012 at 10:38

[GoogleCodeExporter]

increasing limit_ncpus in ./cpyrit/config.py does not have any reliable effect 
but shows mimal increase in PMKs processed

Original comment by joris.la...@gmail.com on 12 Sep 2012 at 10:44