I use used 32 threads per block size in my paralellization of the nested for loop found in the Evolvetimestep function. The results were very good.