[omid20110]

Hello everybody
I want to speed up my OpenFOAM simulations. I have figured out that there are several ways to do this:
1- using gpu;
2- using cluster
3- openfoam cloud

However I have searched the net but couldn't find a good answer to my questions:
1- Is there any gpu solver for sprayFoam?" Does it make sense to use gpu?

2- And also what about cluster? How can I use it? and also does it better than cpu parallel simulation with core i5 Cpu?

3- What is openfoam cloud? How can I use it?

Thanks in Advance

Best regards,
Omid

[Tobi]

Hi,

I don't know why everybody is sending me private messages and ask for help but well,...

• To speed up OpenFOAM simulations you can run the case in parallel and you should renumber your mesh.
• FOAM is programmed in C++ and not in CUDA, hence it is not an GPU architecture software and therefore you can not use the GPU. There should be a way with special keywords to make the C++ sources more/less usable for GPU's but finally it is not a GPU architecture. In addition, the speed up is not that much, forget that kind of idea.
• If you want to speed up the stuff use graphic cards, you really should reprogram everything in CUDA. Maybe there are some open source codes available which deal with that but I have no experience in that.
• Cloud computing should be the same as parallel computing but instead of having your own cluster, you use some cluster shared by some company (some cluster that does not belong to you).

Summing up. FOAM is written C++ and not developed for GPU calculations. Speeding up the simulation can be done with multi core processors using the parallel stuff. Keep in mind that the processor boundary conditions influence your solution and if you split the case into too much subdomains it can happen, that you slow down instead of speeding up based on the fact that the communication between the CPU's is high. Further speed up can be achieved in solving the matrices within a good and fast solver (GAMG, FDIC etc.), compiling the code with special CPU flags (optimization of the code) and of course thinking about your mesh.

Other opinions are welcomed but this is the state of the art in my personal perspective.

[hk318i]

Hi,

As Tobias said, OpenFOAM doesn't work on GPU out-of-box. However, there is a project is trying to do that which I never tried myself (RapidCFD)

Regarding, cloud and clusters, your questions are quite general. There are many cloud computing services, mostly paid, which you can use; Amazon, google, UberCloud, simScale, conself, MS Azure . . . etc.

Best Regards,
Hassan