I should warn you right from the beginning that the algorithm used in GraphEnt is neither the most stable nor the most efficient of those published3 (but it is the one that is the easiest to code). Additionally, for those cases where the calculation includes a known figure-of-merit for the phase angles, GraphEnt is performing additional approximations which although I hope that are generally safe, they do not represent the best that can be achieved with the data.
I should also warn you that the amount of time that the calculation may require depends on the input data quality and there is no a priori guarantee that the given algorithm will converge even if given enough time. Having said that, a 262,144 (=128x64x32) pixels GraphEnt mFoexp(i) map corresponding to a reasonably accurate data set could be calculated in less than 8 minutes of CPU time on a DEC Alpha 1200, a 524,288 (=128x128x32) pixels GraphEnt difference Patterson map for a loosy derivative (which makes the calculation easy) took only 46 seconds on the same machine, and a 2Å (2mFo - DFc)exp(i) synthesis with 3,072,000 (=160x160x120) pixels took 40min. Finally, a 2-0.8Å (2Fo - Fc)exp(i) synthesis with 9,437,184 (=192x256x192) pixels for a 4--helical bundle protein took only 13 minutes on a Pentium 800MHz (but the data were, of course, rather weak at this resolution range).