accelerate the performance of simpleFoam solver using paralution library on GPU

savee · June 1, 2017, 06:13

Hello,

I am trying to accelerate the performance of simpleFoam solver in Openfoam on GPU. To integrate openfoam with gpu, I am using paralution library. I have integrated the simpleFoam solver of OpenFOAM-2.4.0 with paralution_PCG. The solver is compiled successfully. But, when I execute a test case using that compiled solver, I get the increase in execution time as compared with the execution time required to execute a test case using simpleFoam solver (without using paralution_PCG). I have taken grid size of 40*40*40 and executed the test case for different number of gpu cores like 4, 8, 16, 32, 64. Even, I have tried by increasing the grid size to 80*80*80, the same result is there. As paralution library is used to accelerate the performance of the solver, but in my case it is not like that..
What is the problem.. why this is happening?
Can anyone please help me..

khedar · June 12, 2017, 08:45

In HPC you cannot just assume that by increasing the number of cores you will get the desired speedup. It depends on lot of factors, major of which is the memory access. If you don't get the desired speedup there is a bottleneck process (memory access) somewhere.

For e.g read this https://cs.stackexchange.com/questions/12670/does-petsc-really-give-speedup

You have to find out what is taking more time in the solving process. This is called profiling.

manasitiwari · March 13, 2019, 07:18

Hi Savee,

I don't know if you found the solution to your problem but maybe you can solve mine. I am not able to use Paralution with OpenFOAM. You have mentioned in your post that you were able to compile OpenFOAM solver's with paralution library. I was going through Paralution's documentation and they mention a Make/file option. I dont know where to look for it. Can you please help me?

I have been stuck on it for 2 days.

Thanks and Regards,
Manasi

sunagra27 · January 7, 2020, 00:37

Hi Savee,

I am working on chtMultiRegionSimpleFoam and I would like to accelerate using paralution library. I have installed paralution using user manual as suggested. The installation is done well.

But, I am getting a problem while adding, for example "paralution_PCG.C" to Make/files folder and also in options folder. I am getting an error stating the " No rule to make target 'Make/linux64GccDPInt32Opt/".

With respect to this, I know I am missing something in Make file, but I am unable to find the reason fot it. I have also added path to "LD_LIBRARY_PATH".

Please help me out with this as your paralution is working fine.

Thanks and Regards,

Sunag R A.

June 1, 2017, 06:13	accelerate the performance of simpleFoam solver using paralution library on GPU	#1
savee New Member Savita Mane Join Date: Mar 2017 Posts: 4 Rep Power: 9	Hello, I am trying to accelerate the performance of simpleFoam solver in Openfoam on GPU. To integrate openfoam with gpu, I am using paralution library. I have integrated the simpleFoam solver of OpenFOAM-2.4.0 with paralution_PCG. The solver is compiled successfully. But, when I execute a test case using that compiled solver, I get the increase in execution time as compared with the execution time required to execute a test case using simpleFoam solver (without using paralution_PCG). I have taken grid size of 404040 and executed the test case for different number of gpu cores like 4, 8, 16, 32, 64. Even, I have tried by increasing the grid size to 808080, the same result is there. As paralution library is used to accelerate the performance of the solver, but in my case it is not like that.. What is the problem.. why this is happening? Can anyone please help me..

March 13, 2019, 07:18	paralution with openfoam	#3
manasitiwari New Member Manasi Tiwari Join Date: Aug 2018 Posts: 3 Rep Power: 7	Hi Savee, I don't know if you found the solution to your problem but maybe you can solve mine. I am not able to use Paralution with OpenFOAM. You have mentioned in your post that you were able to compile OpenFOAM solver's with paralution library. I was going through Paralution's documentation and they mention a Make/file option. I dont know where to look for it. Can you please help me? I have been stuck on it for 2 days. Thanks and Regards, Manasi

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Compressible Solver Implementation in FSI Library (Foam-extend 3.2)	KarthickRajkumar	OpenFOAM Programming & Development	1	January 4, 2017 06:45
fluent divergence for no reason	sufjanst	FLUENT	2	March 23, 2016 16:08
How to run simpleFoam as an invsicd solver?	tareqkh	OpenFOAM Running, Solving & CFD	2	August 10, 2015 16:58
Working directory via command line	Luiz	CFX	4	March 6, 2011 20:02
compressible two phase flow in CFX4.4	youngan	CFX	0	July 1, 2003 23:32

June 12, 2017, 08:45		#2
khedar Senior Member khedar Join Date: Oct 2016 Posts: 111 Rep Power: 9	In HPC you cannot just assume that by increasing the number of cores you will get the desired speedup. It depends on lot of factors, major of which is the memory access. If you don't get the desired speedup there is a bottleneck process (memory access) somewhere. For e.g read this https://cs.stackexchange.com/questions/12670/does-petsc-really-give-speedup You have to find out what is taking more time in the solving process. This is called profiling.

January 7, 2020, 00:37		#4
sunagra27 Member Sunag R A Join Date: Jul 2019 Location: Bangalore, India Posts: 82 Rep Power: 6	Hi Savee, I am working on chtMultiRegionSimpleFoam and I would like to accelerate using paralution library. I have installed paralution using user manual as suggested. The installation is done well. But, I am getting a problem while adding, for example "paralution_PCG.C" to Make/files folder and also in options folder. I am getting an error stating the " No rule to make target 'Make/linux64GccDPInt32Opt/". With respect to this, I know I am missing something in Make file, but I am unable to find the reason fot it. I have also added path to "LD_LIBRARY_PATH". Please help me out with this as your paralution is working fine. Thanks and Regards, Sunag R A.