The security of multivariate public-key cryptography is largely determined by the complexity of solving multivariate quadratic equations over finite fields, a.k.a. the MQ problem. XL (eXtended Linearization) is an efficient algorithm for solving the MQ problem, so its running time is an important indicator for the complexity of solving the MQ problem. In this work, we implement XL on graphics processing unit (GPU) and evaluate its solving time for theMQ problem over several small finite fields, namely, GF(2), GF(3), GF(5), and GF(7). Our implementations can solve MQ instances of 74 equations in 37 unknowns over GF(2) in 36,972 s, 48 equations in 24 unknowns over GF(3) in 933 s, 42 equations in 21 unknowns over GF(5) in 347 s, as well as 42 equations in 21 unknowns over GF(7) in 387 s. Moreover, we can also solve the MQ instance of 48 equations in 24 unknowns over GF(7) in 34,882 s, whose complexity is about O(267) with exhaustive search.