GPU-Acceleration of Tensor Renormalization with PyTorch using CUDA

Abstract

We show that numerical computations based on tensor renormalization group (TRG) methods can be significantly accelerated with PyTorch on graphics processing units (GPUs) by leveraging NVIDIA's Compute Unified Device Architecture (CUDA). We find improvement in the runtime and its scaling with bond dimension for two-dimensional systems. Our results establish that the utilization of GPU resources is essential for future precision computations with TRG.

Comments: v2: Added computation of critical exponents for three-state Potts and XY model. Comparison tests between CPU and GPU for Ising at T=Tc. Addressed the comments by the anonymous referee. Version matches the one accepted for publication in Computer Physics Communications. v1: 1+15 pages, comments welcome!