The performance of adaptive optics systems for existing as well as future giant telescopes heavily depends on the number of active wavefront compensating elements, the spatial, and the temporal sampling of the distorted incoming wavefront. In a phase-A study for an extreme adaptive optics system for the VLT (CHEOPS) as well as for LINC-NIRVANA a fizeau interferometer aboard LBT with a multi-conjugated adaptive optics system, we investigate how today's off-theshelf computers compare in terms of floating point computing power, memory bandwidth, input/output bandwidth and real-time behavior. We address questions like how level three cache can impact the memory bandwidth, what matrixvector multiplication performance is achievable, and what can we learn from standard benchmarks running on different architectures.