Abstract. Unsteady free-wake solutions of wind turbine flow fields involve computationally intensive interaction calculations, which generally limit the total amount of simulation time or the number of turbines that can be simulated by the method. This problem, however, can be addressed easily using high-level of parallelization. Especially when exploited with a GPU, a Graphics Processing Unit, this property can provide a significant computational speed-up, rendering the most intensive engineering problems realizable in hours of computation time. This paper presents the results of the simulation of the flow field for the NREL Phase VI turbine using a GPU-based in-house free-wake panel method code. Computational parallelism involved in the free-wake methodology is exploited using a GPU, allowing thousands of similar operations to be performed simultaneously. The results are compared to experimental data as well as to those obtained by running a corresponding CPU-based code. Results show that the GPU based code is capable of producing wake and load predictions similar to the CPUbased code and in a substantially reduced amount of time. This capability could allow freewake based analysis to be used in the possible design and optimization studies of wind farms as well as prediction of multiple turbine flow fields and the investigation of the effects of using different vortex core models, core expansion and stretching models on the turbine rotor interaction problems in multiple turbine wake flow fields. IntroductionWind turbine placement in a wind farm for power optimization requires specialized tools that can provide accurate results in short times. These problems have not yet been tackled efficiently by conventional Computational Fluid Dynamics (CFD) codes, which require too long of a computational time to offer any feasible attempt to solve wind farm wake interaction and layout optimization problems. The blade element momentum theory (BEM), an alternative, provides an exceptional performance from the standpoint of computational time. However, attainable accuracy restricts its use only to preliminary sizing and performance analysis of wind turbines. The advantages of both methods, i.e. high accuracy and low computational time, are offered by free-wake panel methods.