This is the first course in a new series on using GPUs in Python. It is introductory and doesn’t require any knowledge besides basic Python. Experience with NumPy is useful, but not necessary.
I will start with the basics of using GPUs on the Alliance supercomputers.
Then I will briefly introduce CUDA and the various libraries that make it accessible in Python.
Finally, I will focus on CuPy, a high-level library that provides a drop-in replacement for NumPy and SciPy. We will see how to convert code from NumPy to CuPy and how to write GPU kernels when existing routines aren’t sufficient. While doing so, we will benchmark executions on CPUs and GPUs and monitor GPUs usage.