Nowadays whenever a complex, parameterised function needs to be optimized, a parameter estimator will be employed that varies the function’s parameters according to some algorithm and evaluates the changes in the result and thereby eventually finds the optimal parameter set. One such estimation process would be the execution of the pattern search as first described by Hooke and Jeeves in 1961. This estimator or optimizer shall be the basis for a new parallel implementation suited for GPU computing using the OpenCL language that will be presented in this work. This solution will be described, discussed and evaluated using triple exponential smoothing as the function to optimize. The goal is to minimise the forecast error by setting the parameters accordingly. Several variants of the proposed algorithm are compared in experiment and for additional comparision they will run against a simple grid search.