Explore techniques for writing custom C and CUDA extensions to optimize GPU computations in this 26-minute conference talk from EuroPython 2020. Learn how to create and seamlessly integrate GPU-accelerated functions with Python codebases through a detailed example of image search on billions of items. Compare methods for developing extensions and bindings in terms of speed and ease of use, and discover how to modify setup.py files for straightforward deployment and sharing of optimized GPU code.