RAPIDS on Kaggle#

Kaggle Notebooks provide hosted Jupyter notebooks with GPU accelerators. The latest Kaggle notebook environment includes RAPIDS libraries such as cuDF and cuML, so you can start using them without a separate installation.

Create a GPU notebook#

  1. Log in to Kaggle or create a Kaggle account.

  2. Select Create and then Notebook. You can also open the new notebook page directly.

    Kaggle Create menu with Notebook selected

  3. In the notebook editor, select Settings > Accelerator, then choose an available GPU option, such as GPU T4 x2 or GPU P100. Starting a GPU session uses your Kaggle GPU quota, so stop the session when you are done.

    Kaggle notebook Settings menu showing the available GPU accelerators

  4. Select Settings > Environment Preferences > Always use latest environment. This makes the notebook use Kaggle’s latest environment, which contains the preinstalled RAPIDS libraries.

    Kaggle notebook Environment Preferences menu with Always use latest environment selected

Test RAPIDS#

Enter the following code in a notebook cell, and select the Run button or press Shift+Enter:

import cudf

gdf = cudf.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
gdf.sum()

The cell should produce the following output:

a     6
b    15
dtype: int64

This confirms that cuDF can use the prebuilt Kaggle environment. You can also import cuml to run GPU-accelerated machine learning workflows.

Next steps#

Try these examples in your Kaggle notebook: