Graphcore develops hardware accelerators for machine learning and AI. A new integration with Paperspace has made it easier than ever to spin-up an IPU in a data science notebook environment.

Getting Started

The first step is to head over to Paperspace and log in to the console. If you haven’t signed up for Paperspace yet, it’s free and takes about 60 seconds.

From the console, we’ll head over to Gradient and create a new project.

Next, we’ll be creating a new notebook using one of the pre-configured Graphcore runtimes. These runtimes automatically download some starter files into your notebook. They also specify the Graphcore IPU machine type and the pre-configured container that already has all the dependencies on it that you need to run workloads on the IPU.

After we start the notebook it will take about a minute to clone the repo and configure the machine.

The machine is now running!

Let’s start running some code!

Fantastic! We’re now up and running with a Graphcore IPU on Paperspace.

If we want to check out more starter projects, Paperspace has provided three different starter runtimes, including HuggingFace, PyTorch, and TensorFlow.

For more information on specific Graphcore IPUs available on Paperspace, be sure to read the docs!