Skip to main content

Quickstart

Top-ups open soon

支付宝 (Alipay) top-ups open soon, and 微信支付 (WeChat Pay) follows once the domain moves to HTTPS. Until then you cannot top up online; write to finance@superheat.cloud with any questions.

This walks the shortest path from a new account to a shell on a GPU. It assumes nothing except a mainland-China mobile number (or an email address) and Alipay.

1. Sign in​

Go to console.superheat.cloud, enter your mainland-China mobile number and select Continue with SMS. You'll get a text message with a six-digit code — there is no password to choose. To use email instead, switch the sign-in card to Email and the code goes to your inbox.

The first time you sign in, a personal organization is created for you automatically. Everything you rent belongs to an organization, and yours is ready immediately.

See Create your account for the details.

2. Add credits​

Superheat is prepaid: to deploy, your balance has to cover at least one hour of the instance — the Total line of the deploy page's cost panel (GPU plus disk) — on top of an hour of the organization's existing instances. A balance merely above zero is not enough.

Open Billing → Add credits (账单 → 充值额度), choose Alipay under Pay with, and pick a ¥50, ¥100 or ¥500 pack, or enter a custom amount between ¥20 and ¥10,000. Select Show QR code (去支付) and scan the code with Alipay. The code is valid for 15 minutes; credits land on your balance as soon as Alipay confirms the payment.

For WeChat Pay, a 发票 (fapiao), and paying on a phone, see Add credits.

Only organization admins can add credits

On your personal organization you are the admin, so this is not a restriction until you join someone else's team.

3. Add an SSH key​

If you want shell access — and you probably do — add a public key before deploying, so it's baked into the instance at launch.

Generate one if you don't have it:

ssh-keygen -t ed25519 -C "superheat"

Then paste the contents of ~/.ssh/id_ed25519.pub into SSH Keys → Add SSH key. See Supported key types for what the form accepts.

4. Pick a GPU​

Open Rent GPUs. Each card is an offer — a slice of a real machine, showing its GPU model, how many GPUs are in the slice, VRAM per card, TFLOPS, storage, location, a reliability score, and the hourly price for the whole slice.

Filter down to what you need, then select an offer. See Browse the marketplace for what every filter does.

5. Deploy​

Selecting an offer raises a selection bar at the bottom of the page; Continue takes you to Configure & Deploy. You choose:

  • A template — start with a system template such as PyTorch (SSH), PyTorch (Jupyter) or CUDA Devel. See Templates.
  • Disk size — from 10 GB up to the machine's own disk capacity, in 10 GB steps. The API's ceiling is 20,000 GB, and a deploy asking for more than the machine has is rejected. Disk is billed separately while the instance is stopped, so do not over-provision out of habit.
  • An SSH key — the one you added in step 3. Leaving it as No key (web terminal only) means no SSH access.
  • A label (optional, up to 64 characters), so you can tell instances apart later.

A cost panel updates as you change these, showing the GPU rate, the storage rate, the total per hour, an approximate per-day figure and your balance. Select Deploy instance.

6. Connect​

The instance goes through creating and then running. The instance page has Connect, Logs and Billing tabs; Connect holds a ready-made command:

ssh -p 40123 root@203.0.113.14

The host part is the machine's public IP: SSH always reaches the machine directly, never through a relay. The port is not 22 — each instance publishes its SSH port from the host's range, and the exact port is assigned at launch. Copy the command rather than reconstructing it.

If your template uses the jupyter launch mode, an Open JupyterLab button appears above the SSH command, already carrying the access token. That mode opens port 22 as well, so both routes work.

7. Stop or destroy when you're done​

This is the step people skip, and it's the one that costs money.

ActionContainerYour diskGPU slotBilling
Stopshut downkeptheld for youstorage only
Destroyremoveddeletedreleasednothing

Stop when you'll be back later today and want your data waiting. Destroy when you're finished. Read Stop vs destroy before you decide — a stopped instance keeps charging.

What next​