Launch an instance

Launch a GPU instance from a catalog offer, then monitor, connect to, stop, and destroy it from the Instances list.

Once you've found an offer on the catalog screen, click Launch to provision it. The Instances list gives you a single view of everything you have running.

Launch an instance

From an offer's row in the catalog, click Launch and choose the container image you want the instance to boot. If you're planning to connect and work interactively, a general-purpose image with CUDA and your usual tooling is a reasonable default; if you're running a fixed job, bake it into the image instead.

1. 1
   Pick an offer

   From the catalog screen, click Launch on the offer you want.
2. 2
   Choose an image

   Select the container image the instance should boot — a general-purpose image for interactive work, or a custom one for a fixed job.
3. 3
   Confirm launch

   The instance appears in your Instances list immediately, and its status updates live as it comes up.

Monitor and inspect

The Instances list shows every GPU instance on your account in one place. Open a single instance to see its full status and connection details, including the host and port you'll need to connect over SSH.

Connect over SSH

Once an instance reports as running, its connection details in the console show the SSH host and port. Connect from your own terminal using the key pair you registered on your account — see SSH keys.

What SSH access practically involves

• A key pair, not a password — instances only accept the SSH key(s) already registered to your account under SSH keys. Add a key there before you launch, or the instance boots with no way in.
• Host and port are per-instance — each instance gets its own connection details, shown on its entry in the Instances list once it reports running. They're not fixed across instances, so re-check them every time you launch a new one.
• Root access on the box — once connected, you have the same level of control as any bare Linux GPU host: install packages, mount data, and run long jobs inside a terminal multiplexer so they survive a dropped connection.
• Nothing survives past the instance — files live on that instance's disk only. Copy off anything you need before you stop or destroy it.

Stop, destroy & clean up

Destroying an instance from the Instances list stops the running GPU immediately so billing ends. Once an instance is stopped, you can also remove its record from the list.