Optimizing CarpetX Performance with Subcycling Integration

In this talk, we introduce a new implementation of subcycling in CarpetX, which will enable the physical thorns to run faster on both CPU and GPU machines. This new subcycling method requires fewer buffer points but achieves higher-order convergence compared to the old method implemented in Carpet. Fewer buffer points improve the scaling performance, and we present the scaling performance on both CPU and GPU machines. We compare the accuracy of the simulation results with and without subcycling and found that they agree quite well.