Learn how to leverage GPUs to improve chip-design quality and make the VLSI design process faster. We will show a GPU-Accelerated global placement engine built on PyTorch that achieved a 40X speedup over multi-threaded implementation and can place a 10M cell design in four minutes. In addition to direct GPU acceleration of design automation software, we'll explain how we apply a deep learning approach to physical design problems, which indirectly leverages GPU. We will illustrate a method that leverages convolutional neural networks and fully convolutional networks to predict design rule checking hotspots during physical design. This deep learning-based approach significantly outperforms other ML approaches such as support vector machine in prediction accuracy.