Setting up managed instances

Thunder Compute provides managed instances that are optimized for performance. The Thunder Compute CLI contains all of the functionality you need to create, edit, use, and delete these instances.

Start/create an instance

• $ tnr create
  • Creates a new Thunder Compute instance and automatically assigns an instance ID
• $ tnr start [instance_ID]
  • Starts an inactive Thunder Compute instance

Stop/delete an instance

• $ tnr stop [instance_ID]
  • Stops a running Thunder Compute instance
• $ tnr delete [instance_ID]
  • Deletes a Thunder Compute instance

Connect to a running instance

• $ tnr connect [instance_ID]
  • Connects you to a running Thunder Compute instance
  • This is a wrapper on SSHing. You can also manually SSH if desired using the IP displayed in $ tnr status

List instances

• $ tnr status
  • Lists all TNR instances and active GPUs associated with your Thunder Compute account
  • Any time a GPU is actively running a job you can check on it here

Running GPU code

To run code on Thunder Compute GPUs, prefix any command with the $ tnr run command. This gives any process direct access to a GPU, without paying when the GPU is idle. You can use $ tnr device cpu to guarantee that you are disconnected from a GPU, however this is not required. If the GPU is idle you will never pay.

Authentication

The Thunder Compute CLI uses the TNR_API_TOKEN environment variable for authentication. This saved token is used to authenticate all operations. You can generate these single-use tokens in the console. Thunder Compute tokens do not expire and you can revoke them at any time. To access Thunder Compute from multiple devices, generate separate tokens for each device.

Storage

Storage on Thunder Compute instances is persistent. By default, instances have 100GB of storage which can be configured with $ tnr create. If you have significant amounts of data in a storage bucket and are constrained to a specific region or cloud provider, contact our support team.