Finding the low-energy spectrum of spin glasses by exhaustive search using CUDA

We present how to find a low-energy spectrum for small (N <50) spin glasses of any structure, using modern GPUs or similar heterogeneous architecture. The presented algorithm performs a full review of all possible system configurations in order to select l << 2^N states with the lowest energy, together with the corresponding energies. During the seminar we will present the results of performance tests of our implementation of the discussed algorithm, we will also show its practical application in verifying the operation of the MPS method.