When GPUs Make Python Slower: Patterns and Pitfalls

This conference talk from PyCon US explores counterintuitive scenarios where GPU acceleration actually slows down Python code. Discover both successful and unsuccessful use cases for GPU acceleration through practical examples. Learn how GPU-accelerated libraries like Numba, CuPy, cuDF, and cuGraph function behind the scenes. Examine specific challenges including string processing limitations, memory transfer bottlenecks, and real-world time-series data processing scenarios. In this 33-minute presentation, gain insights to better evaluate when GPU acceleration is beneficial for your Python projects and when alternative optimization approaches might be more effective, regardless of your experience level with GPU computing.