Recognition of potential for protein-RNA interaction is an important problem in bioinformatics. The solution may present a clue for understanding gene regulation. Formalization of the problem leads to in silico search for a complex motif in the 15-letter UIPAC alphabet in RNA sequences considering their secondary structure. The genetic algorithm with island modification was used to solve the related discrete optimization problem. The algorithm named SWORD was implemented on GPU and CPU. The comparison has shown a significant performance advantage of the GPU implementation. The algorithm was applied for searching RNA-motifs interacting with Hu antigen R (HuR) protein. The result achieved is better than that obtained in the previous work based on fitness-function criterion.