Computational performance and scaling of adiabatic quantum annealing processors

We characterise the recent 128 qubit quantum annealing processor, D-Wave One, through investigation of hardness and scaling of ``time-to-solution'' for several thousand realisations of $\pm J$ spin glass problems, ranging from 8 to 108 qubits in size. We compare statistics of the results to classical- and simulated quantum annealing. Within the processors noise and calibration uncertainties, we find that the results generated by the D-Wave One are statistically indistinguishable from results generated by a simulated quantum annealer while significantly different from those of a classical annealer. An intriguing feature is strong bimodal separation of the instances into two categories: hard and easy. This feature is not observed for the classical annealer. Based on the similarities between the simulated quantum annealer and D-Wave One, we make predictions for the 512 qubit processor, D-Wave Two.