Publication Date

Fall 2021

Degree Type

Master's Project

Degree Name

Master of Science (MS)


Computer Science

First Advisor

Ben Reed

Second Advisor

Adam Kochanski

Third Advisor

Robert Chun


WRF-SFIRE, wild fire spread models, GPU acceleration, physical simulation


WRF–SFIRE is an open source, atmospheric–wildfire model that couples the WRF model with the level set fire spread model to simulate wildfires in real time. This model has many applications and more scientific questions can be asked and answered if the model can be run faster. Nvidia has put a lot of effort into easing the barrier of entry for accelerating applications with their tools to be run on GPUs. Various physical simulations have been successfully ported to utilize GPUs and have benefited from the speed increase. In this research, we take a look at WRF-SFIRE and try to use the Nvida tools to accelerate portions of code. We were successful in offloading work to the GPU. However, the WRF-SFIRE codebase contains too many data dependencies, deeply nested function calls and I/O to effectively utilize the GPU’s resources. We look at specific examples and try to run them on a Titan V GPU. In the end, the compute intensive portions of WRF-SFIRE need to be rewritten to avoid data dependencies in order to leverage GPUs to improve the execution time.