Datashader is a Python library for creating principled visual representations of large datasets. This talk is a case study on the effort to add GPU support to Datashader using the RAPIDS project. In addition to discussing the performance benefits that were attained, we’ll cover a variety of design patterns that were used to maximize code reuse between the CPU and GPU code paths. This talk will be useful for anyone interested in learning about how to incorporate RAPIDS technologies into their own projects.