Use NVIDIA apps
Prior reading: Overview of cloud app types
Purpose: This document provides additional information about using NVIDIA apps in Workbench.
Introduction
Workbench gives you the option to create NVIDIA cloud apps. These apps run inside containers that are built from a base NVIDIA image. You can interact with these apps directly via JupyterLab.
There are two app options: NVIDIA NeMo and NVIDIA Parabricks and CUDA-X Data Science.
NVIDIA NeMo is a scalable generative AI framework that simplifies building, customizing, and deploying AI models. It's purpose-built for LLMs, multimodal models, and speech AI.
NVIDIA Parabricks is a software suite designed for secondary analysis of genomic data, such as read alignment and variant calling. NVIDIA CUDA-X Data Science is a software suite designed to accelerate statistical analyses. Both software suites leverage GPU computing, allowing for faster statistical, genomic, and multiomics analyses.
Be aware
NVIDIA app types can incur higher costs than other Workbench app types because of the VMs being used. Costs will vary based on the GPU type.When you launch your NVIDIA app, Workbench will open JupyterLab in another browser tab.
Installed software
The initial app image for both NVIDIA app types includes the following software:
- CUDA 12.x*
- NVIDIA-SMI 570 family
The image for NVIDIA Parabricks and CUDA-X Data Science apps also includes the following:
- Python 3.10.10*
- JupyterLab 4.4.9*
* These software are not pinned to a specific version. Apps will install the latest version.
The image for NVIDIA NeMo apps includes JupyterLab, but the version is managed by NVIDIA.
Create your NVIDIA cloud app
On the Apps tab of your workspace, select + New app instance to open the Creating app dialog. Select either the JupyterLab (NVIDIA Parabricks and CUDA-X Data Science) or JupyterLab (NVIDIA NeMo) app.
On the Compute options step, you can customize the GPUs, machine type, and data disk size. You'll see compatible machine types based on your selected GPU configuration.
Note
For NeMo apps, the GPU types have limited availability and may not be included in your current quota. If you have any issues launching your NeMo app, contact Workbench support.
You may also need to set up reservations on Google Cloud Platform (GCP) to ensure you have access to specific GPUs and machine types. See Compute resource reservations for cloud apps for more information.
Once the app is created, launch it in JupyterLab by selecting the name of your NVIDIA app on the Apps tab. Note that the app must be running.
Last Modified: 27 October 2025