Hi all, I've been a long-time lurker of the WRF forums for various debugging over the years and decided to finally make an account so that I may be more active within the WRF community!
I have a question that's been pretty difficult to track down within the literature that I hope someone may be able to offer some insight on. My question is:
Are there any estimates out there as to what fraction of total computation time is spent running each of WRF's individual modules? For instance, if total model computation time is about 1 hour, can I get some rough guesses as to how much of that computation time was spent on microphysics, how much was spent on the PBL scheme, how much was spent on convective parameterization (if one is being utilized), etc.
I understand that these values may be variable and subjected to change depending on domain size, grid spacing, and choice of parameterization options. I'm not looking for anything comprehensive, but am interested in hearing if there are any papers out there that discuss this in length or if any users here may have some knowledge on the subject or are able to point me in the right direction. Even something along the lines of "microphysics is the single most intensive WRF module" would be incredibly helpful!
My reason for wanting to know this is that I've been tasked with exploring GPU-accelerated capabilities for WRF. Within past literature, there has been some success in porting over individual WRF modules and dramatically speeding up computation time. But, I'm not interested in speeding up the computation time for a module that makes up a relatively small fraction of the total time spent running the model, and, unfortunately, it's a bit vague how beneficial some of these speed-ups really are.
Any help is much appreciated and thank you in advance!
I have a question that's been pretty difficult to track down within the literature that I hope someone may be able to offer some insight on. My question is:
Are there any estimates out there as to what fraction of total computation time is spent running each of WRF's individual modules? For instance, if total model computation time is about 1 hour, can I get some rough guesses as to how much of that computation time was spent on microphysics, how much was spent on the PBL scheme, how much was spent on convective parameterization (if one is being utilized), etc.
I understand that these values may be variable and subjected to change depending on domain size, grid spacing, and choice of parameterization options. I'm not looking for anything comprehensive, but am interested in hearing if there are any papers out there that discuss this in length or if any users here may have some knowledge on the subject or are able to point me in the right direction. Even something along the lines of "microphysics is the single most intensive WRF module" would be incredibly helpful!
My reason for wanting to know this is that I've been tasked with exploring GPU-accelerated capabilities for WRF. Within past literature, there has been some success in porting over individual WRF modules and dramatically speeding up computation time. But, I'm not interested in speeding up the computation time for a module that makes up a relatively small fraction of the total time spent running the model, and, unfortunately, it's a bit vague how beneficial some of these speed-ups really are.
Any help is much appreciated and thank you in advance!