currently I run bivariate model in GCTA on a cluster at the university to find out the genetic correlation on two traits. on the cluster, it provide two methods to require the paralle environment, the MPI option and OpenMP (share memory) option.
But when I use the MPI option, the maximum cores gcta can use is alway 7, no matter how many cores I assign for it. It seems when using MPI, cluster can assign cores from different node into one job. For OpenMP, everything works fine, if I need 40G memory I can assign 20 cores in one node to the job. But the problem is at our univeristy there are only limited number of nodes (6) with that much of cores (32). The rest of them only have 12 cores. So basically my jobs using 30 cores will never get start, becuase it has very low priority.
So I am writing to ask if "--reml-bivar" in GCTA can combine memory from different node (MPI)?