NVIDIA/cudf

What it solves

It accelerates tabular data processing by moving data manipulation tasks from the CPU to the GPU, significantly speeding up operations on large datasets.

How it works

cuDF provides a suite of libraries that leverage CUDA to process dataframes. It includes a C++ core (libcudf) with Apache Arrow compliant structures, Python bindings (pylibcudf), and a high-level Python API (cudf) that mirrors the pandas API. It also integrates as a GPU engine for Polars and provides a GPU backend for Dask DataFrames.

Who it’s for

Data scientists and engineers who use pandas, Polars, or Dask and need to process massive tabular datasets more efficiently using NVIDIA GPUs.

Highlights

• Pandas Compatibility: Offers a DataFrame library that mirrors the pandas API and a zero-code change accelerator (cudf.pandas) for existing pandas scripts.
• GPU Acceleration: Uses CUDA C++ and Apache Arrow to perform high-speed tabular data processing.
• Broad Integration: Supports GPU acceleration for Polars (via cudf-polars) and Dask (via dask-cudf).
• Ecosystem Support: Powers other high-performance tools like Spark RAPIDS and the Sirius SQL engine.