In this work, we proposed a new algorithm of the tensor renormalization group in four-dimensions, and showed improvement in computation time. We also performed parallel calculations on multiple-GPUs.