Hourly and per-minute GPU rental: how billing works

Who it is for

This kind of billing suits work that does not need a GPU around the clock: fine-tuning a model over an evening, running a render, trying an idea on a different card, hosting a demo for a day. A server for a day or a week works too: it is the same meter, running for longer.

How to start

1. Sign up and top up your balance with cryptocurrency, or through SBP from a Russian bank.
2. Pick a card and a server type in the catalog, on-demand or interruptible: the hourly price is fixed at launch.
3. Start the server with a template or your own Docker image and work; the balance goes down every minute.
4. When you finish, stop the server (the disk is still billed) or delete it (nothing is left to pay for).

Rental price

Questions and answers

What is per-minute GPU rental?

You pay only for the minutes a server with a graphics card is on. The server has a price per hour, charged from your balance by the minute: 40 minutes of work cost 40 minutes.

How does GPU server rental with hourly pricing work?

The hourly price is fixed at launch and does not change while the server runs. There is no minimum term and no prepaid package: money is charged from your balance by the minute.

Can I rent a GPU by the day or for a week?

Yes, though there are no separate plans. A server for a day or a week is billed the same way as one for an hour: for every minute it is on, at the hourly price fixed at launch.

Does a stopped server pay for the graphics card?

No, the GPU is not charged. The disk storage is still billed, and the console shows what it costs before you stop. A deleted server costs nothing.

Can the price change while my server is running?

No. The hourly price is fixed at launch and stays the same while the server is on. New catalog prices apply to servers you start later and to restarting a stopped server: if the price has changed, the console shows the new one before it starts.

How is an interruptible server different from an on-demand one?

An interruptible server costs less, but the host may stop it if another customer bids more; the disk stays. An on-demand server is not interrupted. Both are billed the same way, by the minute.

Guides

• How not to overpay for GPU rental: stopping, disk and trafficWhat is actually billed when renting a GPU per minute: the GPU, the disk while stopped, and traffic. How to stop or delete a server so you do not pay for idle time.
• What interruptible GPU servers are and when to use themAn interruptible server costs less than on-demand, but it can be paused if someone outbids you. Which tasks suit it and how to avoid losing work.
• Which GPU do you need for AI? Pick by task and VRAMA chat with a 14B model needs 12 GB, FLUX and 27–32B models 24 GB, 70B models 48 GB. A task table, GPUs by memory size and live rental prices.

Other ways to rent

• GPU VPS rental: a Linux server with a graphics card and root over SSHGPU VPS rental: a Linux server with root over SSH and a whole NVIDIA card. How it differs from a regular VPS, how to set it up and what it costs, billed per minute.
• Rent a GPU for AI: LLMs, images, video and trainingRent a GPU for AI: run an LLM, Stable Diffusion, ComfyUI or LoRA training on an NVIDIA card. Ready templates, per-minute billing from your balance, SSH access.
• Rent a graphics card online, billed by the minuteRent an NVIDIA graphics card on a server: RTX 3090, 4090, 5090, A100, H100. Starts in minutes, billed per minute from your balance, with SSH and ready AI apps.

Solutions

• Fine-tuning and LoRA training on a rented GPULLM fine-tuning and LoRA training on a rented GPU: how much memory a 7B model needs, QLoRA, Kohya for SDXL and FLUX. Time and cost estimates.
• JupyterLab with a GPU for rent: PyTorch, CUDA and data scienceJupyterLab with CUDA and PyTorch on a rented GPU, billed per minute. Experiments, notebooks and data prep without buying a GPU.
• GPU rendering on a rented server: Blender Cycles and 3D by the minuteRendering in Blender Cycles on a rented CUDA GPU. When renting beats buying, how to start a render over SSH, and what an hour costs.