The RAPIDS 24.12 release introduced several important updates to cuDF:
Availability on PyPI: Starting with version 24.12, CUDA 12 versions of cuDF and associated libraries are available on PyPI, simplifying the installation process and integration into existing Python environments.
Performance Improvements: This release includes optimizations that speed up groupby aggregations and improve the efficiency of reading files from AWS S3, contributing to more responsive data processing workflows.
Support for higher memory requests than GPU: The Polars GPU engine, powered by cuDF, now supports larger memory requests than the GPU through CUDA Unified Memory, enabling the management of datasets that exceed the physical memory of the GPU.
Enhanced Graph Neural Network Training: Improvements have been made to facilitate faster training of graph neural networks (GNNs) on real-world graphs, broadening the applicability of cuDF in advanced machine learning tasks.
