How instances work
An instance is a container with an AMD Radeon GPU attached. You start it from Home in the console, choose what runs in it and how large it is, and it runs until you destroy it or a platform limit stops it.
What you choose at launch
Section titled “What you choose at launch”| Setting | What it decides |
|---|---|
| Deploy type | Notebook gives you JupyterLab and SSH for development. Model API serves a model with vLLM behind an OpenAI-compatible endpoint. See Serve a model. |
| Size | The number of GPUs, CPU cores and memory, and the credits it costs per hour. |
| Image | The software in the container: ROCm, Python, frameworks. See Choose an image. |
| Resource pool | The group of GPU nodes the instance is scheduled on. |
| Workspace storage | Whether /workspace is deleted with the instance or kept. See Keep your data. |
| Mount a model | Model files mounted read-only, at /models for most models. See Keep your data. |
The form shows Deploy type, Size and Image by default. Turn on Customize to see the rest.
What your account can use
Section titled “What your account can use”The sizes, images and resource pools in the form are the ones assigned to your account, so another account may see different options. Being listed means you may request a resource. It does not reserve a GPU.
To see the same list from a terminal, run rcc catalog list. See Radeon Cloud CLI.
One instance per account
Section titled “One instance per account”Each account runs one instance at a time. While it runs, Home shows the instance instead of the launch form. Destroy it before you launch another.
Billing
Section titled “Billing”A running instance spends 1 credit per GPU every hour. The rate is shown on the size you pick, for example 1 GPU · 16 cores · 55GB · 1 credit / hour. The first hour is charged within a minute or two of the instance becoming ready. The instance card on Home shows how long the instance has been up and how many credits it has used.
Your balance has to cover the first hour, and the instance is deleted when your credits run out. See Credits.
Limits that stop an instance
Section titled “Limits that stop an instance”An instance is deleted automatically when your credits run out. It’s also deleted once the Auto-reclaim time shown on the launch form has passed since it became ready, unless the form reads Not enabled. Save results to a persistent /workspace as you go, so a stop doesn’t cost you the work.
SSH is always on
Section titled “SSH is always on”Every instance launched from the console comes up with SSH. Add your public key in Settings before you launch. See Connect over SSH.