Dual Processor Multi Core Parrell Processing Question

Question

Hey Guys
I'm looking for a little clarification on an issue with parrell
processing in LabView. If I have a Dual Processor machine with two 4 core CPU's
will be able to access all 8 cores in the LabView environment. I'm presuming it
can use any cores the operating system can see?   
Thanks for the help,
Tom
Solved!
Go to Solution.

Answer 1

Norbert B wrote:
it is the job of the OS that applications can use all cores if necessarry. So for the application itself, it should make no difference if the system (in hardware) is MultiCPU, MultiCore or even simply HyperThread.....
Norbert 
Its true, but I would like to add my 5 cents here.
Lets say, if you have single loop like
while (true){
//do something
then OS will get no chance to run it in multiple threads. So, you will get max 12,5% CPU load at 8 cores PC or 50% max on dual core PC.
I have dual core PC right now, and lets check it:
So, as we can see - 50% CPU load reached (one core loaded more, but its another story).
Well, if we will use two while loops, then we will get 100 % load:
Of course, if you will need to load all 8 cores, then you should have 8 parallel loops.
Compare BD above with the following:
We have two Array minmax functions, and they independend, but we have 50% only.
Well, you can get also 100% CPU utulization withing single while loop. In th example below you have two SubVI, which called in the same loop:
We have here 100 %. Important, that these VIs should be reenterant!
See what happened if they not reeenterant:
Now a little bit about Vision. Behing of most of the Vision SubVIs are DLL calls. Some Vision functions already optimized for multicore execution. For example, convolution:
On the BD above we have single loop with one SubVI, but both cores are used (because convolute itself already optimized for multi core).
Remember, that not all Vision functions optimized yet. For, example, LowPass still single-threaded (compare this BD with BD above):
Sure, we can utilize multi cores - just perform parallel execution (you have to split image to two parts, then join together and so on):
Remember, that SubVIs should be reeentrant, and all DLL calls should be thred safe (not in UI thread). Also good idea to turn off debugging in such experiments for eliminate additional CPU load.
Another point about 8 cores. As far as I know, LabVIEW (and LabVIEW-based application) will support only 4 cores within one execution system by default (at least prior to LabVIEW 2009). If you need to utulize all 8 cores, then you should add some lines into LabVIEW.ini. Refer to the following thread where you can found more details:
Interpolate 1d slow on 8 core machine
Hope all written above was correct. 
Thank for reading and best regards,
Andrey.
Message Edited by Andrey Dmitriev on 11-27-2009 02:50 PM