WRF Acceleration Options

[lslrsgis]

Dear WRF community，

I am trying to accelerate WRF simulation since processing long-term climate simulation. The factors which might influence the speed are listed for my knowledge:

• Hardware (CPU)
• MPI: serial/smpar/dmpar/
• Compiler
--PGI/intel fortran/GNU
--optimization
• Namelist
--time_Step
--physical options

My questions are:
(1) Someone argues that using Intel/PGI as compiler would be faster that GNU, is it true?
-Intel/PGI compiler + dmpar would be the best choice? Are they free and mpi supported?
-what is the optimization options when compiling?
(2) How to optimize the namelist.input setting for my purpose?
-For example, the time_step is necessary to be 180, and history_interval to be 180s, 60s for domain1&domain2, respectively? (attached namelist.input.01 only for 1 year)
-dfi would be necessary for a long-term climate simulation?

Is there any references for this topic? Thanks in advance.

Siliang

 

[kwerner]

Hi Siliang,

1) Typically Intel and PGI run WRF simulations faster, but they take longer to compile the code, so if you plan to modify a lot of code and need to compile many times, you may not want a compiler that builds slowly; however, if you only plan to compile once and run large runs (as you've indicated), then Intel or PGI may be a better choice. Unfortunately those 2 compilers are not free and you must pay for a license to install them. You can take a look at this talk given at our WRF tutorial for a bit more information on compiling options:
http://www2.mmm.ucar.edu/wrf/users/tutorial/201901/werner_compiling.pdf

2) For namelist best practices, take a look at these 2 web pages:
http://www2.mmm.ucar.edu/wrf/users/namelist_best_prac_wps.html
http://www2.mmm.ucar.edu/wrf/users/namelist_best_prac_wrf.html


Kelly