User Tools

Site Tools


howto:start

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
howto:start [2026/02/20 09:42]
ccrosby
howto:start [2026/02/24 10:42] (current)
ccrosby [Using the Intel compilers]
Line 10: Line 10:
 ===== Tips and Tricks ===== ===== Tips and Tricks =====
 [[howto:TipsAndTricks|Tips & tricks for common tasks.]] :!: ** Please read this.  The information is based on hard won experience. ** :!: [[howto:TipsAndTricks|Tips & tricks for common tasks.]] :!: ** Please read this.  The information is based on hard won experience. ** :!:
 +
 +
 +===== If you need more memory =====
 +As of May 2025, 41 compute nodes have had more memory added, cannibalizing memory modules from nodes that have failed permanently.  These 24-core nodes now have 256 GB of RAM each.  You can access these nodes by including the directive ''mem=250gb'' in your resource request line.  These nodes work with the usual queues such as smp, normal or large.
 +
 +<code>
 +#PBS -l select=1:ncpus=24:mpiprocs=24:mem=250gb 
 +</code>
 +
 +Only use these nodes if you need more than 128 GB of RAM, but less than the 1TB available on the "fat nodes".
 +
  
 ===== Using the Intel compilers ===== ===== Using the Intel compilers =====
Line 56: Line 67:
 export MPIF77=mpiifx export MPIF77=mpiifx
 export MPIF90=mpiifx export MPIF90=mpiifx
 +export I_MPI_HYDRA_BOOTSTRAP=ssh
 </code> </code>
  
-It is not necessary to use a machinefile for multi-node MPI runs.  However, the default bootstrap method using pbsdsh does not work.  It is necessary to set the bootstrap method to ssh.  This is a typical command mpi run command:+It is not necessary to use a machinefile for multi-node MPI runs.  However, the default bootstrap method using pbsdsh does not work, hence the need to set ''I_MPI_HYDRA_BOOTSTRAP''.  This is a typical command mpi run command:
 <code> <code>
-mpirun -np 96 -bootstrap ssh myMPIprogram.exe+mpirun -np 96 myMPIprogram.exe
 </code> </code>
  
  
  
-===== If you need more memory ===== 
-As of May 2025, 41 compute nodes have had more memory added, cannibalizing memory modules from nodes that have failed permanently.  These 24-core nodes now have 256 GB of RAM each.  You can access these nodes by including the directive ''mem=250gb'' in your resource request line.  These nodes work with the usual queues such as smp, normal or large. 
  
-<code> 
-#PBS -l select=1:ncpus=24:mpiprocs=24:mem=250gb  
-</code> 
  
-Only use these nodes if you need more than 128 GB of RAM, but less than the 1TB available on the "fat nodes".+ 
 ===== Transferring data to DIRISA ===== ===== Transferring data to DIRISA =====
 Long-term data storage through [[howto:Dirisa|DIRISA.]]  Long-term data storage through [[howto:Dirisa|DIRISA.]] 
/app/dokuwiki/data/attic/howto/start.1771573321.txt.gz · Last modified: 2026/02/20 09:42 by ccrosby