I’m working on setting up the Weather Research and Forecasting (WRF) model to run on NVIDIA GPUs and would like guidance on the best approach. Specifically, I’m looking for recommendations on:
Compilers: Which compilers support GPU acceleration for WRF?
CUDA Support: What are the required CUDA libraries or dependencies for GPU acceleration?
Installation and Compilation Steps: Are there any detailed guides on how to properly build WRF with GPU support?
Performance Considerations: Any advice on optimizing performance when running WRF on GPUs?
If anyone has successfully set up WRF on an NVIDIA GPU, I’d really appreciate your insights or references to relevant documentation.
You should direct these questions to NCAR, who maintains WRF.
It been quite awhile since I’ve built WRF so don’t know their current state. When I did it, it was straight forward to build and listed as one of the build options. They were using pgfortran at the time, which has since been rebranded to nvfortran.
Thank you for your response, since you’ve built WRF in the past, could you share any documentation or references that you used during your setup? It would be helpful to understand the steps and any challenges you encountered.
Sorry, but my experience is unlikely to be very relevant. I project lead the SPEC CPU 2017 benchmark 521.wrf, so spent much of my time removing the code from the WRF build system and putting in the SPEC CPU benchmark harness.
I did build it from their build system as well a few times, but that was just running their “configure” script and selecting the appropriate option, then running “compile”. It’s straightforward and I didn’t need to refer to any documentation.
You’ll be much better off going to the source and asking the folks at NCAR for advice. They have a user forum as well. WRF SUPPORT