GPU Accelerated Solver of Time-Dependent Air Pollutant Transport Equations

Main objective of this paper is to outline possible ways how to achievea substantial acceleration in case of advection-diffusion equation(A-DE) calculation, which is commonly used for a description of thepollutant behavior in atmosphere. A-DE is a kind of partialdifferential equation (PDE) and in general case it is usually solved bynumerical integration due to its high complexity. These types ofcalculations are time consuming thus the main idea of our work is toadopt CUDA platform and commodity GPU card to do the calculations in afaster way. The solution is based on method of lines with 4th orderRunge-Kutta scheme to handle the integration. As a matter of fact, theselected approach involves number of auxiliary variables and thus thememory management is critical in order to achieve desired performance.From a technical point of view, we have implemented a particularvariant of the A-DE system, where the pollutant concentration istime-dependent. An efficient data handling is primarily based on theexploitation of shared memory blocks and texture caches inside GPUchip. Detailed evaluation of the obtained results is given in thispaper where an astonishing execution speed up of GPU-based solution isdemonstrated in comparison to standard CPU.