The code to be tested (Lare3d) is a Fortran95 code with MPI parallelisation. The default setup from the tar file below requires ~1.5GB of RAM and should have a runtime of less than 10 minutes on a single core of most systems. These tests are purely to assess the performance of a single mode, i.e. do not rely on an interconnect. The only output from the code is to two small files in the Data directory and full-scale MPIIO has been turned off so these tests are purely of computing power. The Makefile may be edited for optimisation but all compiler options, and the compiler used, should be stated in the results. The only other file which needs to be edited is in /src/control.f90 in which subroutine 'control_variables' defines the value of nx_global. This needs to be changed for the tests. The default code must be run without other editing unless this is required to correctly compile and run on your system. If this is needed then all changes must be reported.

1. Deafault code on a single core
2. Default code on 2, 4, 8 cores of the same node
3. Defauls code on all cores of a node if this is greater than 8
4. Edit code to rerun on all cores of the node with nx_global = 128 * number of cores. 
   

Any questions regarding these benchmark test should be addressed to Dr. Tony Arber (t.d.arber@warwick.ac.uk). Note Dr. Arber will be unavailable from 27 August to 10 September. If there are questions during this time these should be addressed to Dr. Matt Ismail in the Centre for Scientific Computing.

The source code is avalable as a tar file UKMHD Benchmark Tests