GPU Acceleration Using CUDA for Computational Electromagnetics

Although initially designed for graphics rendering applications, Graphics Processing Units (GPUs), due to their massively parallel many-core nature, have seen widespread use in accelerating many computationally expensive algorithms in recent years. In particular, GPUs have revolutionized the field of machine learning, enabling training deep neural network models which are multiple orders of magnitude larger than previously possible. Even consumer-grade GPUs, which have become very affordable and can be found in personal workstations (e.g., the NVIDIA GeForce RTX 4090), can achieve upwards of 80 TeraFLOPS computing performance.