GPU Programming

A Carpentries Incubator lesson on programming Graphics Processing Units (GPUs) in Python.

The rendered lesson is available at: https://carpentries-incubator.github.io/lesson-gpu-programming/

Lesson Content

The lesson covers the following topics:

• Introduction: overview of GPU architecture and when to use a GPU
• CuPy: GPU-accelerated NumPy-like arrays
• PyTorch: tensor operations and @torch.compile for general-purpose GPU computing
• Numba: JIT-compiling Python functions for the GPU
• CUDA fundamentals: writing and launching GPU kernels using CUDA and Python
• Memory hierarchy: global, local, shared, and constant memory
• Streams and events: concurrent kernel execution and fine-grained synchronization

Prerequisites

Learners are expected to have:

• Basic Python programming skills
• Familiarity with NumPy arrays

No prior GPU or parallel programming experience is required.

Teaching this lesson?

This material is open-source and freely available. If you are planning to use it in your teaching, please get in touch at training@esciencecenter.nl. We would love to help you prepare and receive feedback based on your experience.

Contributing

Contributions are welcome! Please read CONTRIBUTING.md before opening a pull request or issue.

Current maintainer(s)

• Alessio Sclocco @isazi