Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

WRF parallel run without increasing grid cells number

This post was from a previous version of the WRF&MPAS-A Support Forum. New replies have been disabled and if you have follow up questions related to this post, then please start a new thread from the forum home page.


New member
Dear WRF community,

We are carrying out a WRF MPI run with the following domains dimensions

e_we = 58, 76, 85,
e_sn = 31, 37, 49,

Because the decomposition limit problem, in this case we are limited to use

nproc_x = 5,
nproc_y = 3,

However, we would like to take advantage of the entire processor, i.e using 32 cores. For this, we must rescale domains dimensions so that each task takes 10 grid cells as minimum. But increasing the grid cells number would increase computational time, and this is something unnecessary in our case, in which domains dimensions above mentioned are enough for our propose.

So, how can we use 32 cores in WRF parallel run without increasing grid cells number?

Thank you very much for your attention,
Nast Acosta García,
Harmattan International Atmosphere Researching Group.
You mention that the domain sizes you've chosen are large enough for your purpose. Have you done some considerable testing with these domains? We actually never recommend using a domain with a size smaller than about 100x100 grid cells. Anything smaller does not give large-scale propogations enough space to actually resolve anything. Even if you are only interested in microscale events, the large-scale factors are important for forcing. I would recommend taking a look at this web page that we have put together for best practices for the WPS namelist. It also provides a lot of useful information for creating a "good" domain.

and if you're interested, there is a page for namelist.input, as well:
Dear kwerner,
first at all, thank you for your fast response and the useful information.

Consider now that we have a "good" domain with the following size

e_we < 320,
e_sn < 320,

In this case, is there any way to run WRF model using 32 cores in parallel?
Yes, absolutely! You could even use more for a domain that large. You may find this FAQ helpful, as well. It describes good methods for choosing the right number of processors for a particular domain size:
This clear up my questions!
Just a last one. Minimum decomposed computational patch size, either x-dir or y-dir, is 10 grid cells, but in you affirm

For your smallest-sized domain:
((e_we)/25) * ((e_sn)/25) = most amount of processors you should use

Then, which would be the minimum decomposed computational patch size we should use, 10 or 25 grid cells?

The hard rule is that the absolute minimum decomposed patch size is 10 in each direction. The equation listed:
((e_we)/25) * ((e_sn)/25) = most amount of processors you should use
is more of a "rule of thumb," which means a safe/ideal number that can be played around with for each case. But I guess to answer your question, 10 is the definite minimum.