Since NVIDIA brought GPU-accelerated computing into the mainstream with CUDA, a growing ecosystem of high-level libraries has made it straightforward to run languages such as Python, Julia, and R on GPUs.
RAPIDS cuDF extends this capability to Python DataFrames. It can be used in 3 ways: directly; through a pandas-like API; or via Polars’ lazy API.
In this course, I will introduce GPU computing on the Alliance clusters, briefly review the core concepts of CUDA programming, and compare the different ways to access RAPIDS cuDF. The majority of the course will then focus on using cuDF with Polars, as this approach is both the simplest and the most efficient.