Computing platform for OpenFOAM(Acceleration with GPU vs. CPU)

cagri.metin.ege · August 11, 2016, 15:24

Dear All,

We want to know increase OpenFOAM performance for Direct Numerical Simulation with GPU or/and CPU acceleration on Linux OS;

Which acceleration CPU vs GPU much more efficiency for OpenFOAM solvers? Any reference papers or benchmark study will help to figure out increasing performance of OpenFOAM.
If we choose to build a cluster with multiple CPU processors, which hardware combination getting the best performance, some emerging technology, and new product efficiency support each other, i.e. how fast cache communicate between motherboard and CPU portal through current products.
Is OpenFOAM support GPU acceleration or which library is more successful in that class?
Considering OpenFOAM support MPI library, Can OpenFOAM support hybrid architectures (shared and distributed memory together) cluster or just run pure distributed memory architectures?
Which type parallelism of OpenFOAM has coarse-grained, fine-grained parallelism or another way/combine parallelism?

Kind Regards

hxaxtma · August 15, 2016, 04:59

I recently published a performance study on HPC Systems (CPU), maybe this helps you:
https://www.researchgate.net/publica...rmance_Systems

and here a small report regarding CPU vs GPU:
http://vratis.com/blog/sample-page/

cagri.metin.ege · August 17, 2016, 03:04

Dear hxaxtma,

Thank you for replied. I surprise that Intel compiler did not show significant difference GNU, Moreover, It's performance is considering for finer meshes.

As we know in CPU parallelism, increase number of cores not linearly increase performance (Amdahl's law). Of course, we sacrifice it for CPU parallelism. What about GPU parallelism/acceleration?

Lots of documents point out GPU acceleration is significant to improve performance compared the CPU. Is there any drawback for GPU acceleration?

Best Regards.

August 11, 2016, 15:24	Computing platform for OpenFOAM(Acceleration with GPU vs. CPU)	#1
cagri.metin.ege New Member Cagri Metin Join Date: Jul 2013 Location: İzmir/Turkey Posts: 14 Rep Power: 13	Dear All, We want to know increase OpenFOAM performance for Direct Numerical Simulation with GPU or/and CPU acceleration on Linux OS; Which acceleration CPU vs GPU much more efficiency for OpenFOAM solvers? Any reference papers or benchmark study will help to figure out increasing performance of OpenFOAM. If we choose to build a cluster with multiple CPU processors, which hardware combination getting the best performance, some emerging technology, and new product efficiency support each other, i.e. how fast cache communicate between motherboard and CPU portal through current products. Is OpenFOAM support GPU acceleration or which library is more successful in that class? Considering OpenFOAM support MPI library, Can OpenFOAM support hybrid architectures (shared and distributed memory together) cluster or just run pure distributed memory architectures? Which type parallelism of OpenFOAM has coarse-grained, fine-grained parallelism or another way/combine parallelism? Kind Regards

August 15, 2016, 04:59		#2
hxaxtma Senior Member Join Date: Jan 2014 Posts: 179 Rep Power: 12	I recently published a performance study on HPC Systems (CPU), maybe this helps you: https://www.researchgate.net/publica...rmance_Systems and here a small report regarding CPU vs GPU: http://vratis.com/blog/sample-page/ cagri.metin.ege likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[Resolved] GPU on Fluent	Daveo643	FLUENT	4	March 7, 2018 09:02
Superlinear speedup in OpenFOAM 13	msrinath80	OpenFOAM Running, Solving & CFD	18	March 3, 2015 06:36
Star cd es-ice solver error	ernarasimman	STAR-CD	2	September 12, 2014 01:01
Hardware spec (cpu & gpu)	SamuelB	Hardware	0	February 2, 2014 16:38
OpenFOAM 13 Intel quadcore parallel results	msrinath80	OpenFOAM Running, Solving & CFD	13	February 5, 2008 06:26

August 17, 2016, 03:04		#3
cagri.metin.ege New Member Cagri Metin Join Date: Jul 2013 Location: İzmir/Turkey Posts: 14 Rep Power: 13	Dear hxaxtma, Thank you for replied. I surprise that Intel compiler did not show significant difference GNU, Moreover, It's performance is considering for finer meshes. As we know in CPU parallelism, increase number of cores not linearly increase performance (Amdahl's law). Of course, we sacrifice it for CPU parallelism. What about GPU parallelism/acceleration? Lots of documents point out GPU acceleration is significant to improve performance compared the CPU. Is there any drawback for GPU acceleration? Best Regards.