What is Superheat Cloud?
Superheat Cloud rents NVIDIA GPUs by the second. You pick a machine from the marketplace, choose what should run on it, and a few moments later you have a container you can SSH into or open JupyterLab on.
There is no quota request, no instance-type paperwork, and no monthly commitment. You add prepaid credits, and they draw down while your work runs — and, more slowly, while a stopped instance holds its disk.
How it fits together
- The fleet is a pool of GPU machines Superheat operates. Each machine publishes its CPU, RAM, disk, network and region, plus a reliability score.
- An offer is a rentable slice of one of those machines: 1, 2, 4 or 8 GPUs, priced as a unit. The GPU model, VRAM per card and TFLOPS belong to the offer.
- A template describes what to run: a container image, a launch mode, ports and environment.
- An instance is an offer plus a template, actually running.
What makes it different from a VM host
Three things are worth internalizing early, because they explain most of the behavior you'll meet later.
You rent a slice, not a machine. An 8-GPU host might be published as one 8× offer, or as separate 1× and 4× offers. The price attaches to the slice, not to each GPU, so a 4× offer is quoted as one hourly rate.
Stopping is not the same as destroying. Stopping an instance shuts the container down but keeps your slot and your disk — and keeps billing you for the storage. Destroying releases everything. This is the single most common source of surprise on the bill, so it has its own page.
Billing is per second, settled every minute. There is no hourly rounding. A sixty-second run costs sixty seconds. A background tick settles usage roughly once a minute, which is also what enforces the automatic safety stops when your balance runs out.
What you need
- An email address. Sign-in is passwordless — a six-digit code, or Google/GitHub SSO.
- A payment method, to add prepaid credits. There is no post-paid invoicing.
- An SSH public key, if you want shell access. You can add one at any time.