Get started with TIR Instances

Launch a TIR Instance on an hourly plan, open JupyterLab in your browser, and run a first command in it. You need a TIR account with a project, and credit to run the instance; for how much, see How much credit do I need to launch an instance?

Create the instance

1. In TIR, open your project. In the side menu, under Compute, select Containers (Instances).
2. Select Create Instance.
3. On the Pre-built tab, choose the image for the framework you work with, such as PyTorch, TensorFlow, or Transformers. These images open JupyterLab. Leave the image on its first version.
4. Keep the name TIR fills in under Instance Name, or type your own.
5. Under Select Computing Resource, choose what the instance runs on:
   • For a GPU, stay on the GPU tab and choose a GPU, such as NVIDIA H200 SXM with 141 GB of GPU memory. Then pick a size under Configuration. A GPU marked Inventory not available is fully booked right now, so choose another.
   • For a CPU-only instance, switch to the CPU tab, choose C3, and pick a size.
6. Under Instance Pricing, keep On-Demand. TIR then bills only the minutes the instance runs.
7. Leave Workspace at 30 GB. You can make it larger later.
8. Leave Network & Security and Advanced Settings as they are. JupyterLab opens in your browser, so you can skip the SSH key.
9. Select Launch Instance. If your account still needs its billing details, TIR asks for them first; complete that step, and the launch goes ahead.

The new instance is Waiting in Manage Instances while TIR finds capacity and starts the container. This takes about two minutes. When the status changes to Running, the instance is ready.

Open JupyterLab

In Manage Instances, select the instance’s App URL. JupyterLab opens in a new browser tab. If your browser blocks the new tab, allow pop-ups for the TIR site in your browser and select the link again.

Run your first command

In JupyterLab, open a new notebook with the Python 3 kernel. On a GPU instance, type this in the first cell and run it:

!nvidia-smi

The output lists each GPU in the instance with its memory and use, and the CUDA Version that its driver supports. On a CPU instance, run !python3 -V instead, which prints the Python version.

Keep your files in /home/jovyan

The instance’s workspace is the folder /home/jovyan, so save your notebooks and data there. Files in the workspace stay when the instance restarts, stops, or starts. Everything outside it goes back to the image’s contents when the container restarts, so packages you install with pip or apt must be installed again after a restart. For where each kind of data lives, see Where your files live on an instance.

Stop the instance when you finish

An On-Demand instance is billed for its plan until you stop or delete it. To stop it, open the instance’s actions menu in Manage Instances, select Stop Instance, tick the confirmation, and select Stop. Once it is Stopped, billing for its plan stops and your workspace is kept. Only the workspace above 30 GB is still billed. To start it again later, see Stop, start, and restart an instance.

What to do next

• Create an instance explains each choice, including committed and Spot billing, storage, SSH keys, start scripts, and add-ons.
• Connect to an instance over SSH to work in a terminal from your own machine.
• Mount datasets and file systems on an instance to bring in data that lives outside the instance.

If the instance stays on Waiting or JupyterLab does not open, see Troubleshooting TIR Instances.