AMD ROCm-DS#
2025-11-12
3 min read time
The AMD ROCm™ Data Science Toolkit (ROCm-DS) is an open-source collection of GPU-accelerated libraries designed to empower data scientists, engineers, and researchers to build high-performance data science applications and machine learning workflows on the ROCm platform. Built upon the core ROCm foundation, ROCm-DS provides a unified, efficient, and scalable environment for end-to-end data science acceleration.
ROCm-DS is a fork of the RAPIDS® open-source project from NVIDIA, extended and optimized for AMD GPUs. It enables users to accelerate both new and existing data science workloads, executing intensive applications with larger datasets at exceptional speed. With ROCm-DS, you can build pre- and post-processing applications for AI models, create big data processing workloads, or accelerate existing data science pipelines with minimal effort.
ROCm-DS delivers a cohesive set of libraries that target every stage of the data science lifecycle, from data ingestion and transformation to graph analytics, mathematical computation, and vector search. Each component is optimized for GPU performance while maintaining user-friendly interfaces compatible with existing data science frameworks and APIs.
The ROCm-DS toolkit includes the following components:
hipDF – A GPU-accelerated DataFrame library offering fast and scalable tabular data manipulation, aggregation, and transformation. hipDF enables high-performance preprocessing, feature engineering, and ETL workflows essential for modern data pipelines. It also supports the acceleration of many existing Pandas workflows with little to no code changes.
hipGRAPH – Leverages GPU acceleration to process and analyze complex graph structures and networks with speed and precision. hipGRAPH supports diverse graph algorithms—such as centrality, traversal, similarity, sampling, and labeling—and integrates seamlessly with hipDF DataFrames across the ROCm-DS ecosystem.
hipMM – The HIP Memory Manager library provides advanced GPU memory management utilities such as efficient allocation, pooling, and data movement to support the various libraries that form part of ROCm-DS.
hipRAFT – Provides a foundational layer of reusable GPU-accelerated primitives for data science and machine learning, including clustering, dimensionality reduction, and statistical operations. hipRAFT serves as the computational backbone for higher-level data science and AI applications.
hipVS – A GPU-accelerated vector search library containing a variety of high-performance approximate and exact nearest-neighbor and clustering algorithms. hipVS integrates seamlessly with DataFrames offered through hipDF, to enable support for a wide variety of data science workloads.
Note
The hipGRAPH libraries are in an early access state. Running production workloads with these libraries is not recommended.
The ROCm-DS organization is open and hosted at ROCm-DS/.
ROCm-DS documentation is organized into the following categories: