Temperature difference using different number of cores

Discussions centered around methodologies for high performance computing.

Temperature difference using different number of cores

Postby dm712 » Thu Nov 07, 2013 12:09 pm

Hello WRF users,

I have been running WRF with the sf_urban_physics=3 (BEP-BEM) with 1 core or 16 cores. These two simulations gave different results for wind and temperature. The temperature is the most notable difference with a 2K higher temperature when using only 1 core. I am using a domain of 20*20 cells each 45km*45km. May be the issue is that my domain is too small and when using 16 cores the tasks are not divided properly?

Did anyone ever experience this issue? Should I be using only 4 cores or something like that, such that the divided domain is at least 10*10?

Thanks a lot in advance for your help.

Dasaraden MAUREE
dm712
 
Posts: 5
Joined: Mon Sep 30, 2013 3:45 pm

Re: Temperature difference using different number of cores

Postby ronbeag » Thu Nov 07, 2013 12:29 pm

In the WRF user manual, or maybe it's the online FAQ, it says that you should chose the number of processors so that you domain is decomposed into squares (north_south * east_west) of no less that 15 * 15. So if your domain is 20*20 and you use 16 processors, you've got squares of 5*5 for each processor, which is much too low. The extra communication between the processors will slow the model down much more than any gain from increasing the number of CPUs. Personally I usually go for squares of about 20*20 per processor, as the scaling in WRF is certainly not linear. So I would only use one processor for you domain. As to why you get higher temperatures with more processors, i can't explain that, but would be interested to hear from anyone who can.
ronbeag
 
Posts: 12
Joined: Thu Jul 26, 2012 12:12 pm

Re: Temperature difference using different number of cores

Postby dm712 » Thu Nov 07, 2013 2:27 pm

Thank for your reply "ronbeag"!
This is what I also figured. Running with one core would have been way too slow... So now I am pushing it may be a little by running with 4 cores. Results are the same as with one core, so I am running my simulations with this setting. Although I still don't know why the temperature would be greater between the simulations. I guess communications between the proc. are disturbed

Das
dm712
 
Posts: 5
Joined: Mon Sep 30, 2013 3:45 pm


Return to High Performance Computing

Who is online

Users browsing this forum: No registered users and 1 guest

cron