The use of deformable mirrors for the compensation of the effect of the atmospheric turbulence in Adaptive Optics (AO) systems requires a very fast computation of the adequate commands with the information provided by wavefront sensors. Providing accurate results with minimum latency is vital to achieve optimum performance. Conventional CPUs, GPUs and also FPGAs have been successfully used in real-time control for AO at IAC for a number of projects (EDIFISE, AOLI, AOconFPGA, GTCAO). Based in this experience, a comparative description is made in this paper, pointing out advantages and drawbacks of each solution as seen by each of the projects.