We have some more doubts.
Our system is a cray system. So how can we compile in a cray system?
If we don't compile mpi how are we supposed to use more than one node? And
each node is having only one p100 card
And without openmp will it use all the cpu cores in the gpu node?
Please clarify the doubts.
On Thursday, December 12, 2019, David A Case <david.case.rutgers.edu> wrote:
> On Wed, Dec 11, 2019, Sruthi Sudhakar wrote:
> >
> >We have been trying the GPU installation of AMBER 18. We encountered the
> >following error. Please suggest any possible solution.
> >
> >Configure command : ./configure -crayxt5 -noX11 --skip-python -cuda -mpi
> >-openmp gnu
>
> Oh my goodness, you have an amazing number of options: both -crayxt5 and
> gnu, both -cuda and -mpi and -openmp. Start with just -cuda and gnu.
> Are you sure you need a parallel cuda build?
>
> >
> >if (NC_defineRemdIndices(ncid, remd_dimension, remd_indices_var_id,&
> >Error: Missing actual argument for argument 'ismremd' at (1)
>
> What version of AmberTools do you have? If you are using AmberTools19,
> then AmberNetcdf.F90 should be in the AMBERHOME/src/pmemd/src folder:
> check that this is the case. If you are not using AmberTools19, then
> use update_amber --upgrade to upgrade.
>
> Vinicius, Dan, someone: in the master development branch, pmemd is still
> using AmberNetcdf.F90 from AmberTools. I think we should make pmemd
> more isolated from AmberTools, even at the cost of some code
> duplication. But I'd like someone who understands all this REMD restart
> business to figure out the best thing to do.
>
> ...thx...dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Dec 11 2019 - 20:00:02 PST
