[mm66]

Originally Posted by edaymo

Hi, MJ,

Thanks for the update. That was a clever solution. I am honestly not sure why the mapped T BC it's working when you include the turbulence model library ("libturbulenceModels.so"), since this BC is not supposed to be compiled in RapidCFD. But nonetheless, well done for getting it working.

However, I don't have a specific idea for your error. Could be a lot of causes, from the grid to a bad initial condition for your turbulence parameters to something RapidCFD (vs. OpenFOAM) related.

Does the same exact case run fine in OpenFOAM v2.3.1, or does the same problem occur in OpenFOAM? If yes, then it's more a general OpenFOAM issue. If no, then it's something related to RapidCFD in particular. This feedback might help with the determination of the root cause of your problems.

Best regards,

Eric

Hi Eric,


First, thanks a lot for your prompt response, I really appreciate it


Initially I was very happy to see it running, then I found it was not done yet...! (learning to be more patient...)


Yes, I just exactly did the same thing that you mentioned. I ran the exact same case with OpenFOAMv1811 (few changes here and there due to the differences between the versions but nothing meaningful was different). It ran perfectly fine with OpenFOAMv1811 so there should be something which needs attention in RapidCFD...


Thanks,
MJ

[edaymo]

Hi, MJ,

That's good that you tested your case with OpenFOAM-v1812. It is highly likely that the same case will run with OpenFOAM v2.3.1, but if you have the ability to test with OpenFOAM v2.3.1, you can test the closest version of OpenFOAM to RapidCFD. Nonetheless, your OpenFOAM v1812 test probably rules out the grid, for example, as the root cause.

From there, you could start deconstructing your case to figure out where the issue is. Examples:

- Will your case run with a constant wall T instead of the mapped T boundary condition?

- Is it feasible to run your case with fixed density (you're modeling water, considering rho is ~1000 kg/m3 and it seems to vary with temperature)? Or... with laminar flow ?

In other words, I recommend changing parameters until you figure out exactly what is causing the errors. Offhand, I am not sure what the problem could be, in part because you are using some features I never tested with RapidCFD (e.g., mapped T BC and possibly more).

Another option is to take a known working OpenFOAM tutorial for chtMultiRegionFoam and getting that to work with RapidCFD.

I wish you good luck with the troubleshooting process.

Best regards,

Eric

[mm66]

Originally Posted by edaymo

Hi, MJ,

That's good that you tested your case with OpenFOAM-v1812. It is highly likely that the same case will run with OpenFOAM v2.3.1, but if you have the ability to test with OpenFOAM v2.3.1, you can test the closest version of OpenFOAM to RapidCFD. Nonetheless, your OpenFOAM v1812 test probably rules out the grid, for example, as the root cause.

From there, you could start deconstructing your case to figure out where the issue is. Examples:

- Will your case run with a constant wall T instead of the mapped T boundary condition?

- Is it feasible to run your case with fixed density (you're modeling water, considering rho is ~1000 kg/m3 and it seems to vary with temperature)? Or... with laminar flow ?

In other words, I recommend changing parameters until you figure out exactly what is causing the errors. Offhand, I am not sure what the problem could be, in part because you are using some features I never tested with RapidCFD (e.g., mapped T BC and possibly more).

Another option is to take a known working OpenFOAM tutorial for chtMultiRegionFoam and getting that to work with RapidCFD.

I wish you good luck with the troubleshooting process.

Best regards,

Eric

Hi Eric,


Thank you very much for your valuable input.

Since I already had OpenFOAMv1812 on my system, I quickly used it to check if it runs there just fine. But I am in the process of getting v2.3.1 for a more meaningful comparison

I will follow your input to get to the bottom of this...

Thanks again and happy holidays


Regards,
MJ

[Dcn]

Hi ,
Can you kindly tell the step by step procedure on how to
Install the rapid CFD , the steps that you have followed to install the rapid CFD
Thanks

[edaymo]

Hello,

RapidCFD installation steps have been documented at https://github.com/Atizar/RapidCFD-dev/issues/93.

Also, some recent experiences with newer versions of CUDA are posted at https://github.com/Atizar/RapidCFD-dev/issues/92.

Hope this helps.

Best regards,

Eric

[klausb]

What would be a good RapidCFD benchmark case?


All the cases from https://github.com/TonkomoLLC/RapidCFD-Tests are now running on my AMD RX 6700M (10Gb RAM) but they require changes to run on a recent OpenFOAM implementation.


Maybe someone is willing to adapt e.g. the pitzDaily case so it can be run on a recent 4 or 8 core CPU to see where we are performance wise?!

[edaymo]

Hi Klaus,
Very nice to hear that the RapidCFD cases are running on your AMD GPU. Well done.
As to CPU versions: The RapidCFD version was adapted from the OpenFOAM 2.3.x pitzDaly tutorial: https://github.com/OpenFOAM/OpenFOAM.../les/pitzDaily


More modern versions of this tutorial are:

OF-11: https://github.com/OpenFOAM/OpenFOAM...luid/pitzDaily

OpenFOAM v2312: https://develop.openfoam.com/Develop...ref_type=heads

Good luck with the comparisons.

Best regards,
Eric

[klausb]

@Eric

Can you suggest another benchmark example that allows an apples-to-apples GPU to multi-core-CPU comparison?

LES pitzDaily uses different equations oneEqEddy vs kEqn (which runs 30x faster on the CPU but it could be the case size or the equation that makes the difference) and damBreak-rapidCFD won't run on OpenFOAM 2312. damBreak-rapidCFD runs 2.9x faster on my RX 6700M than on the GeForce GTX 980M used in the original "benchmark" but I would have expected a factor of 6x as the fp64 performance of my GPU is 6x higher and the case is rather large. BUT really important would be to compare performance based on a large case to a current 8 core CPU.

OpenFOAM2.3.1 is sooo.... outdated, that it can't be installed on a recent Linux distribution as it doesn't meet recent c++ standards, something that starts affecting RapidCFD, too.

[edaymo]

Hi Klaus,

Maybe old OF 2.3.1 cases can be run with docker (older version Ubuntu that can install OF 2.3.1 from the ppa package archives).

Also newer versions of OF seem to be incrementally faster with each release for the same case (this is my qualitative experience; I do not have data to back up this statement).

Yes, RapidCFD is starting to show its age... thanks for your contributions on the RapidCFD-dev site to help it compile with newer versions of compilers!

Cheers,
Eric