We explore the possibilities to accelerate simulations in computational fluid dynamics by additional graphics processing units (GPUs). By examining some examples of stationary incompressible flows from the industrial practice we demonstrate that the potential speedup obtained by deploying GPU accelerated linear solvers alone is limited if standard segregated algorithms are used. However, recently presented velocitypressure coupling algorithms are an attractive alternative to these segregated algorithms. We present an efficient AMG solver for the coupled linear system of mixed elliptichyperbolic character and show that the GPU-accelerated version of this linear solver accelerates the velocitypressure coupling scheme by almost 50 % compared to a competitive CPU implementation. Compared to standard segregated methods, the computing time of the whole simulation is reduced by up to 75 %.