What is serverless GPU, and how is it billed?
Serverless GPU means you send work and the platform finds a GPU for it. You pay while your container runs, including its start, and nothing once it has scaled to zero.
Every row on rent.dev's board links its source. GPU pods run on your own provider key, and that provider bills you.
How a request gets a GPU
You package your model as a container image or a function. When a request arrives, the platform starts a copy of it on a free GPU. More requests start more copies, up to a limit you set, and when requests stop the copies shut down.
What the meter counts
The meter usually runs from the moment a copy starts until it stops. That includes the cold start, when the image is pulled and the model is loaded into GPU memory.
It also includes the idle window: the time a copy stays warm after the last request, waiting for the next one. Some platforms add a charge for the CPU and memory that come with the GPU.
Cold starts and the idle window
A long idle window keeps a copy warm, so the next request is fast, but you pay for the wait. A short window saves money and makes more requests wait for a cold start. Big models take the longest to load, so they feel this trade the most.
Serverless or a pod
A pod is a whole GPU machine that runs until you end it. Its hourly rate is lower, but it bills while it sits idle. Bursty traffic favours serverless; a GPU that is busy all day favours a pod.
Where rent.dev fits
rent.dev's GPU API starts pods, not serverless workers. Its price board lists serverless rates beside pod rates, so you can compare both before you choose, and its sandboxes run the code around your model.
Questions
What is a serverless GPU provider?
A company that runs your container on its GPUs only when there is work, starts more copies as requests grow, and stops them when traffic ends. You pay for the time your copies hold a GPU.
How is serverless GPU billed?
By the time your container holds a GPU, usually counted in seconds. Most platforms count the cold start and the warm idle window as well, so a short idle window and a small image keep the bill down.
Is serverless GPU cheaper than renting a pod?
For bursty traffic, usually yes, because the meter stops when traffic stops. For steady work the higher hourly rate adds up, and a pod you keep busy often costs less over a day.
Does rent.dev offer serverless GPUs?
No. rent.dev starts GPU pods through one API, on your own key at the GPU provider rent.dev connects to, and that provider bills you for them. Its price board still lists serverless rates, so you can weigh them against pod rates.
Agents can start at rent.dev/llms.txt, which lists every route.