GPU environments

With GPU environments, you can reduce the training time needed for compute-intensive machine learning models you create in a notebook. With more compute power, you can run more training iterations while fine-tuning your machine learning models.

GPU environments are currently available to users of paid Watson Studio Cloud plans in the Dallas region. See the Watson Studio pricing plans.

Default environment definitions

Watson Studio offers a default GPU environment definition that you can use to quickly get started with Python and GPU in Watson Studio tools without having to create your own environment definition.

Name Hardware configuration
Default GPU Python 3.7 1/2 x NVIDIA TESLA K80 (1 GPU)
Default GPU Python 3.6 1/2 x NVIDIA TESLA K80 (1 GPU)

GPU environment definitions

To use a GPU environment with a different hardware specification than provided by the default GPU environment definition, you must create a new GPU environment definition where you can specify a hardware size up to 2 x NVIDIA TESLA K80 (4 GPU).

You must have the Admin or Editor role within the project to create an environment definition.

To create a GPU environment definition:

  1. From the Environments tab in your project, click New environment definition.
  2. Enter a name and a description.
  3. Select the GPU environment configuration type.
  4. Select the hardware configuration. Select the size dependent on the complexity of the model operations and the number of model training iterations you’d like to perform.
    • 1/2 x NVIDIA Tesla K80 (1 GPU using 1/2 of a NVIDIA graphics card)
    • 1 x NVIDIA Tesla K80 (2 GPU using 1 NVIDIA graphics card)
    • 2 x NVIDIA Tesla K80 (4 GPU using 2 NVIDIA graphics cards)
  5. Select the software version:
    • Default Python 3.6 GPU

    The environment definition details are displayed, including how many capacity units are consumed per hour for the chosen hardware size. You can change your hardware settings by hovering over the setting.

The GPU environment definitions include a variety of pre-installed open source libraries. If you want to add your own custom libraries, you can create a customization. See Customizing environment definitions.

Using GPU environments in notebooks

After you have created a GPU environment definition, you can select to run your notebook in that environment at the time you create the notebook. GPU environments are available for the Python notebook language only.

In a project, you can create more than one notebook with the same GPU environment definition. Every notebook kernel runs in the same runtime instance in this case and the resources are shared.

You can edit an existing GPU environment definition after a notebook is created; however, to use the changed environment configuration, you must restart the runtime.

You can restart a GPU runtime from the Environment tab when you click the Notebook Info icon if the notebook is open in edit mode.


The following limitation exists:

  • Different machine learning libraries such as TensorFlow, XGBoost and PyTorch are pre-installed when the GPU runtime is started, which you can use to build and run machine learning models in Watson Studio. However, if you plan to deploy your models to the IBM Watson Machine Learning service, you can only build models that use TensorFlow v1.15 or PyTorch 1.2 models with ONNX.
  • GPU environments for notebooks are available in the Dallas IBM Cloud service region only.
  • The number of GPU environment runtimes you can have active at any time can’t exceed the number of GPU units in your cluster.

Next steps