This project used CUDA on Nvidia K80 to price an Autocall structure. Now 8.192m simulations spent only 0.5s while they spent about 2 minutes on CPU!