NVIDIA Brev Integration#
Overview of Workbench and Brev#
Brev is NVIDIA’s platform for launching cloud instances.
Using a Brev-managed cloud instance as a Workbench location significantly simplifies the process of connecting to remote locations.
Brev provides Cloud GPUs#
It’s a paid service with a click-thru UI/UX to launch GPU instances from different cloud providers
It has a CLI that you install locally to simplify connecting to remote machines via SSH
Workbench connects easily to Brev instances#
Workbench comes pre-installed on Brev provisioned instances so you don’t need to install it
You run a single command in the Workbench CLI to connect to a remote location on Brev
Workbench automatically configures the remote location using SSH information provided by Brev
The Brev integration is simple#
Set up the prerequisites
Create an instance on Brev
Run a Workbench CLI command
Start working!
Integrating Brev and Workbench#
Prerequisites#
AI Workbench installed locally
Brev CLI installed locally (see Brev documentation here)
Authenticated Brev CLI (see here for details)
Note
On Windows, you must install the Brev CLI in the NVIDIA-Workbench
WSL distribution, however there
is a complication because sudo
access is blocked in that distribution.
Get a Cloud Instance on Brev#

Log in to Brev, go to the GPU tab, and click Deploy an Instance
Select VM mode on the next screen. Do not select the Jupyter option
Then select the type of GPU you want, and then select the configuration and provider
Configure the disk size, give the instance a unique name, and click Deploy to launch the instance

Use the Workbench CLI to connect#
This feature is currently only implemented in the CLI. Once you’ve connected using the Workbench CLI, the location will appear in the Desktop App and you can use it like any other location. You cannot connect the remote location until it is running.
Run
brev list
to verify the instance is running and get its name,<instance-name>
Run the Workbench CLI command to connect a remote location using the Brev instance name
nvwb create context --brev-instance-name <instance-name>
Follow the prompts to confirm the connection
Go the Main Locations page in the Desktop App and refresh with
ctrl+r
The Brev location should appear as
brev-<instance-name>
FAQs#
How do I install the Brev CLI in the NVIDIA-Workbench
WSL distribution?#
This is complicated by the fact that sudo
access is blocked in that distribution.
However, there is a simple workaround.
Enter the distribution as root from a Windows shell or PowerShell:
wsl -d NVIDIA-Workbench -u root
Once you have entered as root, you can install the Brev CLI with the usual command for Ubuntu:
bash -c "$(curl -fsSL https://raw.githubusercontent.com/brevdev/brev-cli/main/bin/install-latest.sh)"
You may be prompted to authenticate the CLI to your Brev acount. DO NOT DO THIS WHILE RUNNING AS ROOT
Instead exit the shell session, open up the
NVIDIA-Workbench
WSL terminal, and then authenticate to Brev.
Is Brev a cloud provider?#
No. Brev is more of a cloud-aggregator. It provides a simple UI/UX to launch GPU instances from different cloud providers.
Is Brev free?#
No. Brev is a paid service where you pay for the GPU instances you use.
Why is sudo
blocked in the NVIDIA-Workbench
distribution?#
It significantly streamlines the installation process for WSL, since it skips you setting up a username and password
Having a single user in the distribution,
workbench
, eliminates the large number of edge cases that would be caused by users choosing their own user name
Getting around the sudo
block is simple: Enter as root from a Windows shell or PowerShell: wsl -d NVIDIA-Workbench -u root
I know I created a Brev instance, but brev list
doesn’t show it. What’s wrong?#
This is usually because the CLI is pointing at the wrong Brev org.
Get the org name displayed in the top right of the Brev console,
<org-name>
Set the Brev CLI to use that org with
brev set <org-name>
Then, list the instances again with
brev list