cuML
NVIDIANVIDIA's GPU-accelerated machine-learning library in the RAPIDS suite, with scikit-learn-style estimators for clustering, regression, classification, dimensionality reduction, nearest neighbors and time series - more than fifty algorithms, including GPU versions of UMAP and HDBSCAN. Its cuml.accel mode runs existing scikit-learn, UMAP and HDBSCAN code on the GPU with no code changes, and it scales across GPUs and nodes with Dask.
Openness
5 medium confidence- license
- Apache-2.0(OSI)
- source
- public(github.com/NVIDIA/cuml)
- core features withheld
- no — no paid edition of cuML found
Apache-2.0 and built from the public repository, which moved from the rapidsai account to NVIDIA's. NVIDIA describes it as an open-source library, and its enterprise software page does not list cuML among paid components. It needs an NVIDIA GPU to run.
- https://raw.githubusercontent.com/NVIDIA/cuml/main/LICENSE recorded 2026-09-27
LICENSE is the Apache License, Version 2.0, "Copyright 2018 NVIDIA CORPORATION".
- https://raw.githubusercontent.com/NVIDIA/cuml/main/README.md recorded 2026-09-27
README: "NVIDIA cuML is an open-source CUDA-X Data Science library for GPU-accelerated machine learning."
- https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2Fcuml recorded 2026-09-27
Repository record for NVIDIA/cuml, not archived and not a fork, license apache-2.0, last pushed 2026-09-24.
- https://ungh.cc/repos/NVIDIA/cuml/files/main recorded 2026-09-27
Repository tree of 1,206 paths with no ee, enterprise, commercial or pro directory.
- https://www.nvidia.com/en-us/data-center/products/ai-enterprise/ recorded 2026-09-27
NVIDIA AI Enterprise product page; cuML and RAPIDS are not named among its paid components.
Adoption
3 medium confidenceMonthly PyPI downloads of the cuml-cu12 and cuml-cu13 wheels, summed. RAPIDS is also installed through conda and NVIDIA's own package index, which are not counted, so this is a floor.
- https://docs.rapids.ai/install/ recorded 2026-09-27
RAPIDS install page offers conda, pip and Docker; "RAPIDS pip packages are hosted by NVIDIA, some are also available on the Python Package Index (PyPI)."
- https://pypistats.org/api/packages/cuml-cu12/recent recorded 2026-09-27
last_month 274,395 downloads of cuml-cu12
- https://pypistats.org/api/packages/cuml-cu13/recent recorded 2026-09-27
last_month 3,371 downloads of cuml-cu13
Capability
3 high confidencecuML covers much of scikit-learn's catalog and runs it on GPUs, often many times faster, which is a difference of speed and scale rather than of how much of the task it does. Choosing and tuning the model stays with the user.
- https://docs.rapids.ai/api/cuml/stable/ recorded 2026-09-27
Docs landing: cuML "supports 50+ algorithms across all major machine learning categories, including clustering, regression, classification, dimensionality reduction, and time series analysis", with multi-GPU and multi-node support and cuml.accel for running scikit-learn code on GPUs.
Verified 2026-09-27