Jupyter Notebooks

It is extremely common for data scientists and researchers to use Jupyter notebooks for exploratory data analysis. Rafay GPU PaaS provides a turnkey experience for users with a "Notebook as a Service" type experience. Rafay also provides a number of profiles for notebooks that allow users to spin up environments preloaded with the required libraries and frameworks.

Create Notebook¶

To create a notebook, access the Developer Hub and navigate to the home page. The page provides options to create and manage notebooks, which are predefined configurations designed to deliver interactive environments for tasks such as data analysis, visualization, and machine learning development. On the Developer Hub home page, users can either click on View All to access the Notebooks page or click on New Notebook to create a new notebook. Users can also click on the Notebooks menu on the left to directly access the Notebooks page.

New Notebook¶

To create a new notebook,

Select Notebooks from the menu on the left of the console
Click on New notebook
Select a suitable notebook service profile that suits your requirements

Once the profile is selected, provide the required details. If pricing for the selected profile is configured in Global Settings by the Org Admin, a monthly estimate will be displayed.

Provide a Name for the notebook (adding a Description is optional)
Select the desired Workspace from the dropdown list

Now, specify the GPU resources you would like to provide the notebook. Select from the provided options:
- Number of GPUs
- GPU Model/Type (i.e. node type)

A notebook profile maps to a notebook pre-installed and pre-configured with the required frameworks, libraries and software add-ons that match the profile. Profiles allow the data scientist to start using the notebook right away instead of wasting time trying to install and configure all the software on top of the notebook.

Profile	Description
Minimal	Basic libraries and frameworks only
Data Science	With common libraries for data science
Spark	With libraries required for Spark
Tensorflow	With Tensorflow libraries
Tensorflow with CUDA	With Tensorflow and CUDA libraries
PyTorch	With PyTorch libraries
PyTorch with CUDA	With PyTorch and CUDA libraries

Select the profile you would like to use with your notebook from the provided dropdown list.

Click on Deploy to launch the notebook service

It can take a few minutes for the notebook and associated software components to be deployed and ready for use.

Info

Users can deploy multiple notebooks on an instance. The only constraint is whether the underlying instance has the resources required for all the notebooks.

The notebook initially displays a status of In Progress. Upon successful deployment, the status updates to Success

Use Notebook¶

Once a notebook has been successfully deployed onto a compute instance, the user is presented with the URL for the notebook and a token (authentication credentials) to securely access the notebook. To access the notebook, the user can either click on the URL or copy/paste it into a web browser.

They need to then provide the access token as a credential before they can access the Jupyter notebook.

View Notebook¶

Clicking on the notebooks menu will list of all the notebooks the user has access to. Note that notebooks may span different workspaces and different instances. To view details about a specific notebook, users just need to click on the name of the notebook.

Notebook Metrics Dashboard¶

The Metrics Dashboard provides real-time visibility into resource utilization for Jupyter Notebook instances. This dashboard helps track performance and monitor system behavior while notebooks are running.

From the Notebooks list, locate the required notebook, click on the ellipses (three dots) on the far right and select View Metrics. This opens the Notebook Metrics Dashboard for the selected instance.

Dashboard overview¶

The Metrics Dashboard provides real-time and historical visibility into resource usage for the selected notebook instance. System-level metrics and GPU-level metrics (when applicable) are displayed as time-series charts, along with summary values for current, peak, and average utilization.

System charts are displayed at the top of the dashboard.
GPU panels appear only when the notebook instance is provisioned with GPUs.
A list of all GPUs is shown at the bottom of the dashboard, with expandable sections for each GPU to view detailed metrics.

These charts provide instance-level visibility into CPU, memory, and storage utilization.

CPU Utilization: Displays CPU usage over time for the instance.
- Shows current, peak, and average CPU utilization.
- May display Committed vs Usage, indicating reserved CPU limits vs actual consumption.
- Useful for identifying sustained high CPU load or sudden spikes in demand.
Memory Utilization: Shows memory usage over time.
- Indicates current, peak, and average memory consumption.
- Committed vs Usage highlights reserved memory capacity compared to real usage.
- Helps assess memory pressure and identify increasing memory trends.
Storage Utilization: Displays storage usage of the notebook’s associated volume.
- Shows current, peak, and average storage consumption.
- Helps determine data growth and future storage sizing requirements.

GPU metrics (when GPU(s) are present)¶

When GPUs are attached to the notebook instance, GPU usage and performance metrics are displayed. Each GPU is listed individually and can be expanded to view detailed charts.

GPU Utilization: Percentage of GPU processing capacity used over time.
GPU Memory Copy Utilization: Percentage of GPU memory transfer operations (device-to-device or device-to-host) over time.
GPU Temperature: Shows the operating temperature of the GPU and this can help identify thermal throttling conditions.
GPU SM Clocks: Displays the Streaming Multiprocessor (SM) clock frequency over time.
GPU Memory Clocks: Shows the GPU memory clock frequency trends.
Framebuffer Memory Used: Indicates GPU framebuffer memory actively allocated for workloads.
Framebuffer Memory Free: Shows remaining available GPU framebuffer memory. This is useful for evaluating workload fit, model sizing, or concurrency limits.
GPU list view

A list of GPUs is presented below the metric panels. * Each GPU entry shows the GPU identifier and model (for example, NVIDIA H100 80GB). * Expanding a GPU entry displays the full set of GPU metric panels for that specific GPU. * Enables comparison across GPUs in multi-GPU configurations.

Notes: * GPU charts appear only if the instance is backed by a GPU SKU. * If the instance is stopped or not running, charts may display empty data or no activity. * System charts and GPU charts refresh periodically and display time-series trends.

Delete Notebook¶

To delete a notebook, users should click on the ellipses on the far right of the selected notebook and select delete.

Info

Once deletion has been initiated, it cannot be stopped or reversed. Users can create a new notebook if required.