Brute-forcing spin-glass problems with CUDA