NVIDIA RAPIDS

+ Gradient

How to use Gradient and

NVIDIA RAPIDS

together

Harness the power of NVIDIA RAPIDS and Paperspace Gradient

RAPIDS is a collection of open source libraries from NVIDIA that provides machine learning and deep learning toolsets optimized to run on GPU. The goal of RAPIDS is to make it easy to harness GPU parallelism for accelerated processing and training tasks.

RAPIDS projects

RAPIDS projects include the following:

  • cuDF - a pandas-like dataframe manipulation library for GPU
  • cuML - a collection of classical ML libraries accelerated for use on GPU
  • cuGraph - a NetworkX-like accelerated graph analytics library
  • cuSpatial - a GPU-accelerated spatial and trajectory data management and analytics library
  • cuXFilter - a framework to connect web visualizations to GPU accelerated crossfiltering

For the full list of RAPIDS projects, check out RAPIDS on GitHub.

RAPIDS Advantages

RAPIDS has a number of advantages:

  • GPU parallelism for tasks like loading large data chunks into memory
  • Increased GPU utilization to get the most out of your local or cloud hardware
  • Reduced ETL, training, and optimization time during the ML lifecycle

Getting started with RAPIDS

Getting started with RAPIDS on Paperspace is easy. When you create a new notebook you should see the NVIDIA RAPIDS tile in the Recommended Runtimes. After you select the RAPIDS runtime, select a GPU instance and start your notebook!

To get started with RAPIDS, create a notebook using the RAPIDS tile in the Paperspace console.

By default this tile will pull the workspace located here: https://github.com/gradient-ai/RAPIDS.git.

If you would like to pull a different workspace into the RAPIDS container, we invite you to toggle the Advanced Options and enter an alternate workspace.

Toggle Advanced Options if you'd like to change the default workspace

Once you've started up the notebook, you should see a number of examples waiting for you to try!

You can quickly try a demo tutorial like the NYC Taxi Spatial Notebook!

If you'd like to use a full version of JupyterLab, you can always swap over to a full JupyterLab instance by toggling the JupyterLab button in the sidebar.

JupyterLab is available via the Jupyter button in the sidebar of the Gradient IDE

You should now be up and running with NVIDIA RAPIDS on Paperspace Gradient.

If you need any help, be sure to read the docs and if you get stuck, feel free to ask for help.