running cuda-enabled cp2k on multiple nodes with aprun/HPC

[Ada Sedova]

Hi,

I just built cp2k 4.1 using DBCSR and CUDA_PW on a cray HPC system with one 
K20X GPU per node. The single aprun call seems to always launch 3 apruns as 
can be seen with ps, and one always stops very quickly while the others 
seem to keep running. I am testing in an interactive qsub, but it seems 
this is nonstandard behavior and may be problematic in a batched setting. 
And at any rate, this does not seem like correct mpi behavior based on 
other programs I have run.

I wonder if the build was not completely successful? I'm testing with 
H2O-32.inp from the tests/benchmark directory but this also happened with 
C.inp.

Thanks,

Ada
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170912/bc20b366/attachment.htm>