CROCO and GPU

parallelization
1

Hi,

We would like to test the CROCO code for GPU. We have a machine with three NVIDIA GPUs and we were able to compile the code setting nvfortran in the jobcomp.

How do we tell the run_croco.bash to use one or two GPUs?

Any help is appreciated, thanks!

2

Hi Andres,

Thank you for the info. I am also trying to compile croco for GPU but I am getting the error below. The compilation seems to reach the end when all “.o” files are copied but at that point I get the error:

"
/usr/bin/ld: MPI_Setup.o: in function mpi_setup_': MPI_Setup_.f:(.text+0x130): undefined reference to acc_get_num_devices_h_’
collect2: error: ld returned 1 exit status
gmake: *** [Makefile:204: croco] Error 1
"

I see that you mention you used “nvfortran”. Is this a requirement? Is there a simple guide including requirements for compiling croco for GPU?

My GPU is an Nvidia RTX 2000 Ada. Is there a list with compatible GPU cards that have been tested?

Thanks
Konstantinos

3

Hi Konstantinos,

I have a NVIDIA RTX 5000.

Did you used the NVIDIA compiler, nvfortran? It is mandatory, I understand.

4

Thank you Andres. I will try with the nvfortran then.

5

You need to download and install the nvidia HPC sdk toolkit

( High Performance Computing HPC SDK | NVIDIA Developer)

activate the OPENACC key in cppdefs.h and set FC to nvfortran in jobcom.

This will make use of only one GPU.