# What you are renting

Lesson 1 of [GPU cloud foundations for buyers](https://gpuclouds.fyi/academy/gpu-cloud-foundations.md), 12 minutes.

A GPU cloud rents you access to GPUs over the internet. You pay for the time you use, or for a term you commit to. This lesson names the words you will see on a quote.

## The device is not the whole offer

- A device model, such as H100 or A100, does not give the memory variant or the host configuration.
- GPU memory is the device memory that your workload can use. Host memory is separate.
- NVIDIA lists the A100 with 40 GB and 80 GB of GPU memory, so the model name alone does not settle the memory.
- Interconnect describes the links between devices or hosts. It can affect a distributed job. Test it with your own workload.

## Words for capacity

- **Quota**: permission to request a resource. It is not always reserved capacity.
- **Reservation**: it can be a capacity promise, a spending commitment or both. Ask which promise is in your contract.
- **On-demand**: you request capacity as you need it, under the provider's terms.
- **Spot**: the provider can interrupt this capacity. AWS describes its own spot model. It sends an interruption notice two minutes before it stops, hibernates or terminates an instance, on a best effort basis. Do not apply those exact rules to another service.
- **Serverless**: the service manages allocation. It does not mean that startup is instant. It does not mean that every supporting resource is in the GPU rate.

## Check the region you need

A public device catalogue does not reserve capacity. Check that the device is available in the region you need. Ask for the quota and the allocation terms before you plan a launch.

On this site, the place labels Online and Global are directory labels. They are not data centre locations. Read each listing's compute regions field.

## Sources

- GPUClouds guide: GPU cloud terms for buyers
- GPUClouds guide: Choose a GPU cloud for your workload
- AWS documentation, Spot Instance interruptions and Spot Instance interruption notices
- NVIDIA, A100 Tensor Core GPU product page

Also on the [web page](https://gpuclouds.fyi/academy/gpu-cloud-foundations/what-you-are-renting).