
ResultsApproachWe split the overall work of the 26Queens Puzzle into subproblems by preplacing the queens within the first six leftmost columns. Accounting for the symmetry about the center horizontal, the initial column only needed to be placed half way. The resulting total was doubled to comprise the explored preplacements and their symmetric images. This division yielded exactly 25,204,802 subboards that have been explored by our solver instances. ResultsTotal Solution Count22,317,699,616,364,044
overall solutions. SubtotalsYou may download the subtotals
of all of the explored 25,204,802 subboards. An initial overview is given by the table on the subboards obtained from the preplacement of the two leftmost columns. ProgressThe figure to the right depicts the computational progress over time. It clearly shows the periodic computation speed as well as its acceleration by out iterative design improvements and the joining of our sponsor Signalion. It also reveals an outage of our main performer FPGAs in the middle of May. The 13 clearly convex periods in the computational progress correlate with the placements of the queen in the leftmost column. These placements were essentially processed in sequential order. It appears that the complexity of the exploration of a subboard grows as the queens of the preplacement are located closer to the board center. Most notably, the determined subtotals follow the same trend. Performance and EnergyThis figure illustrates the relative performance of a few devices we employed for our computation especially also in contrast to a few highperformance generalpurpose CPUs. You should also note that we ran eight of the Virtex4 LX160 devices from a single 400 Watts PC power supply. The power consumption of this system was measured to be 110 W at the wall plug. The delivered system performance matches the one of 8*22 2.5 GHzQuadCore systems, each of which draws 180 W from the wall plug. Thus, a QuadCore system spends 288 times at much energy at the solution of one subproblem when compared to the FPGA setup. Using processors with fewer cores would even increase this ration as the load of the system peripherie is shared among fewer parallel computations. 
SponsorsContactProf. Rainer G. Spallek Thomas B. Preußer Bernd Nägel
