Performance Difference Between ipp Versions

[adem_metin_cali]

Hi all,

i use two algoritms (Foreground/Background Segmentation Using Histogram Model and Mixture of Gaussians) to obtain foreground and i use 3 different ipp versions:First one is ipp 6.0.0.062,second one is ipp 7.1.1.119 and last one is ipp 8.0.0.083.I expected more performance  with increased versions but i saw that ipp 6 is best,then ipp 8 and last one is ipp 7.What is the reason of this situations????

[Igor_A_Intel]

Hi Adem (is it correct to use Adem as your name?),

could you provide an output from ippcvLibraryVersion or at least an info on which cpu you have, what linking do you use - static or dynamic, performance difference you've faced with (in numeric form), OS, ia32/x64, etc. Currently nothing is clear from your post - no any information to help you or to provide an answer.

regards, Igor

[Roman_T_]

adem metin c. wrote:

Hi all,

i use two algoritms (Foreground/Background Segmentation Using Histogram Model and Mixture of Gaussians) to obtain foreground and i use 3 different ipp versions:First one is ipp 6.0.0.062,second one is ipp 7.1.1.119 and last one is ipp 8.0.0.083.I expected more performance  with increased versions but i saw that ipp 6 is best,then ipp 8 and last one is ipp 7.What is the reason of this situations????

Hi, Adem!

I had a very similar problem while migrating from ipp 6 to ipp 7. In my case ipp had a problem with internal threads (ipp<->OpenMP interaction).

I set number of threads to 1.

ippSetNumThreads(1);

After this modification ipp7 became as fast as ipp6.

Regards,

Roman

[adem_metin_cali]

Igor Astakhov (Intel) wrote:

Hi Adem (is it correct to use Adem as your name?),

could you provide an output from ippcvLibraryVersion or at least an info on which cpu you have, what linking do you use - static or dynamic, performance difference you've faced with (in numeric form), OS, ia32/x64, etc. Currently nothing is clear from your post - no any information to help you or to provide an answer.

regards, Igor

Hi Igor,

First of all thanks for your interest.I have a i7-Q720 and 8 GB ram.I used only one video to measure performance that is 768x576 dimension and ,2 minutes 30 seconds.I used three different ipp versions:IPP 6.0.0.062,IPP 7.1.1.119 and IPP 8.0.0.083.I used Boost library to measure performance.And average performance informations are below:(Hour:Minutes:Seconds.Fractions)

IPP 6.0.0.062=>Average Calculation Time Per Frame=00:00:00.101847(Dynamic Linkage)

IPP 7.1.1.119=>Average Calculation Time Per Frame=00:00:00.125313(Default Linking Method)

IPP 8.0.0.083=>Average Calculation Time Per Frame=00:00:00.112640(Default Linking Method)

[SergeyKostrov]

>>IPP 6.0.0.062=>Average Calculation Time Per Frame=00:00:00.101847(Dynamic Linkage) >> >>IPP 7.1.1.119=>Average Calculation Time Per Frame=00:00:00.125313(Default Linking Method) >> >>IPP 8.0.0.083=>Average Calculation Time Per Frame=00:00:00.112640(Default Linking Method) A couple of more questions: What C++ compiler and what options did you use? Was it the same C++ compiler for all 3 cases?

[adem_metin_cali]

Hi Sergey,

First of all thanks for your interest.I used visual studio 2012 ide for all 3 cases.I did not use any special options.I set  few options for some C\C++ and Linker Options for boost library.

Adem Metin CALI

[adem_metin_cali]

Hi Roman,

First of all thanks for your interest.I examined your post and your problem was about multithrad library according to the intel ipp users guide.But my problem is different because i use always single threaded library.So i don't need to use ippSetNumThreads(1);.

Adem Metin CALI