GPU Implementation of a Gas-Phase Chemistry Solver in the CMAQ Chemical Transport Model

Area of Science:

• Atmospheric chemistry and physics
• Computational science and engineering
• Environmental modeling

Background:

• The Community Multiscale Air Quality (CMAQ) model is crucial for simulating atmospheric phenomena.
• Gas-phase chemistry in CMAQ presents a significant computational challenge due to complex differential equations.
• Existing computational methods limit the speed of detailed atmospheric simulations.

Purpose of the Study:

• To accelerate the numerical integration of gas-phase chemistry within the CMAQ model.
• To implement a graphics processing unit (GPU)-accelerated solver for CMAQ's CHEM module.
• To assess the performance gains of GPU acceleration on air quality simulations.

Main Methods:

• Migration of CMAQ's Rosenbrock solver from Fortran to CUDA Fortran for GPU utilization.
• Development of CMAQ-CUDA, leveraging the Compute Unified Device Architecture (CUDA).
• Testing the CMAQ-CUDA implementation with standard chemical mechanisms (RACM2, CB6R5, SAPRC07).

Main Results:

• CMAQ-CUDA demonstrated substantial speedups in chemistry time steps.
• Simulations with RACM2, CB6R5, and SAPRC07 required only 51%, 50%, and 35% of the original simulation time, respectively.
• The study confirms CMAQ's suitability for GPU acceleration.

Conclusions:

• GPU acceleration offers a viable solution to the computational bottleneck in CMAQ's gas-phase chemistry.
• The novel Rosenbrock solver implementation in CMAQ-CUDA effectively reduces computational burden.
• This advancement enhances the efficiency of atmospheric simulations and air quality modeling.