Add a Brev Instance#

Overview#

NVIDIA Brev lets you rent cloud instances from multiple providers.

Use Brev if you don’t have your own remote systems or you need to work with scalable GPU systems. It requires a paid account to provision a broad spread of GPU instances as well as CPU only instances.

To add a Brev instance you just need to connect the Brev integration and provision an instance.

The Desktop App can add any running instances you have on Brev, and it will boostrap the install on the instance as part of the process.

Depending on the instance type, the Desktop App will be able to stop and start it.

Brev has stoppable and non-stoppable instances, where the difference is based on features of the underlying cloud provider. Newer cloud providers or aggregators may not have stoppable instances.

AI Workbench can manage the stop and start for stoppable instances. Non-stoppable instances must be terminated in the Brev console to halt charges.

On Windows, you must do the full local install in order to use the Brev integration.

AI Workbench currently uses the Brev CLI to work with the platform. The Brev CLI requires WSL2 on Windows, and you need to do the full local install to setup this up properly for AI Workbench.

Key Concepts#

NVIDIA Brev: A paid NVIDIA platform that lets you select GPU cloud instances from different cloud providers.
Stoppable Instance: Ability to stop a cloud instance to halt charges with the ability to restart it again. Disk writes persist between starts. Capability varies by cloud provider.
Non-Stoppable Instance: Inability to stop an instance and requires termination to halt the instance. Instances is permanently deleted and all disk writes are lost.
Activate/Deactivate: Connecting/disconnecting to an AI Workbench location and starting/stopping the Workbench Service on that server. Automatically starts or stops a Brev instance that’s a location.

Add a Running Brev Instance#

Step One: Create a Brev account, connect the integration and create a cloud instance.

Follow the guidance here: Brev Integration
Go to GPUs > Create New Instance
Select a GPU type from the horizontal scroll menu
Select a GPU instance configuration and provider from the available cards
Enter a number, disk-size, in Configure and Deploy > Disk Storage
Enter a name, location-name, in Configure and Deploy > Name Instance
Click Configure and Deploy > Deploy to launch the instance

Step Two: Add the running Brev instance as a location.

Go to Locations Manager > Add Remote Location (top RHS) > Get Cloud Instance
Click Next
Select your Brev org from the dropdown at top right
Select the instance with name location-name you created in Step One.
Click Add Location

Step Three: Once it is added, activate the remote location and work in it.

Select Location Manager > Location Card (brev-location-name)
When the location window opens, get to work.

Managing Brev Instances#

Use the Brev documentation for further guidance and discovery.: You can find it here.
Stoppable instances: AI Workbench can stop and start the instance for you.: When you close a stoppable location, the Desktop App asks whether to Stop Instance or Keep Instance Running. Stopping the instance halts charges and preserves your disk writes for the next session.
Non-stoppable instances: AI Workbench cannot stop the instance directly.: When you close a non-stoppable location, the Desktop App asks whether to Terminate Instance in Brev or Keep Instance Running. Selecting Terminate Instance in Brev opens the Brev console where you can terminate the instance to stop charges. Terminating a non-stoppable instance permanently deletes it and all disk writes.
Check the Brev console if a session ends unexpectedly.: If the Desktop App shuts down or the SSH connection is interrupted, the instance may remain running. Verify the instance state in the Brev console to avoid unexpected charges.