Cards for small GPU services
RTX 306012 GBfrom $0.11/h
Tesla T416 GBfrom $0.055/h
RTX 309024 GBfrom $0.241/h
RTX 409024 GBfrom $0.308/h
Live prices: the cheapest free server in the catalog, updated automatically
A GPU VPS on gpu.nz is a Linux server with root access over SSH and a whole NVIDIA card, with the driver and CUDA installed. You work with it as with a regular VPS, but it is built differently: it is a container, not a virtual machine, and you pay per minute while the server runs.
You need such a server when a regular VPS lacks a graphics card: to run a language model or an image model, fine-tune a model, run a render, or serve an API on a GPU. You get the whole card with its memory, and the usual server work — apt, pip, git, tmux — is the same as on any Linux VPS.
You log in over SSH, install packages and start a process — from the outside it is a VPS. But the product is different, and it is better to know how before you rent.
| Regular VPS | GPU server on gpu.nz | |
|---|---|---|
| Graphics card | Usually none | A whole NVIDIA card, driver and CUDA installed; 1 to 8 GPUs per server |
| What is inside | A virtual machine with its own kernel | A Linux (Ubuntu) container, no kernel of its own |
| System | Linux, often Windows too |
In short
A GPU VPS on gpu.nz is a Linux server with root access over SSH and a whole NVIDIA card with 12 GB of memory or more, from $0.055 per hour per GPU. It is a container with no Windows and no nested virtualization, billed per minute; for a round-the-clock service take an on-demand server, not an interruptible one.
One hour of a GPU server: about $0.055
Estimate at the current per-GPU hourly price. Billed per minute, the price is fixed at launch.
Prices and availability come from the live catalog and update automatically
| Nested virtualization | Sometimes | No |
| Access | root over SSH | root over SSH, the key is added in the console |
| Billing | Usually a monthly plan | Per minute from your balance, the hourly price is fixed at launch |
| Hourly price | Low | Noticeably higher: you pay for the graphics card |
| Websites and mail 24/7 | A good fit | Not what it is for: port links are locked with the rental token |
Add an SSH key
Create a key and paste its public part into the Keys section of the console before you rent: the key is chosen when the server launches.
ssh-keygen -t ed25519Connect
The address and port are on the server page, and a button copies the command.
ssh -p <port> root@<address>Install packages
System packages come from apt, Python environments from venv, libraries from pip. Take the PyTorch build for the CUDA version that nvidia-smi shows.
apt update && apt install -y git tmux htop
python3 -m venv ~/venv && source ~/venv/bin/activateRun long jobs in tmux
The session survives a dropped connection, and tmux attach -t work brings you back to it.
tmux new -s workKeep files on the server's disk
While the server exists, the disk keeps your data, including after a stop. Copy important results to your own machine anyway, with scp or rsync.
This is a container, so services managed with systemctl, as on a regular VPS, usually do not work: run long processes in tmux. The guide SSH on a GPU server covers keys, connecting and copying files in more detail.
Yes, if the service needs a graphics card. A typical case is your own language-model API: the vLLM template starts an OpenAI-compatible API on port 8000, the rental token is the key, and your app only needs the address and the key. The guide on Ollama and vLLM as an API shows how to connect it.
For a permanent service take an on-demand server, not an interruptible one: the host may stop an interruptible server when another customer bids more. And work out the cost up front: multiply the hourly price by the hours in a month. A server with a graphics card running around the clock costs noticeably more than a VPS without one. An API for a 7–8B model needs 12–16 GB, larger models need 24 GB; the calculator works out the memory.
Stopping ends the GPU charge, but the disk keeps your files and is billed as storage until the server is deleted. The console shows what storage costs before you stop. If you no longer need the server, delete it: then nothing is charged. The page on per-minute rental explains how the money is counted.
Watch your balance
When the balance runs out, servers are deleted together with their disks. While your funds run low we warn you by e-mail and in Telegram, but keep a reserve and copy important files to your own machine in advance.
RTX 306012 GBfrom $0.11/hTesla T416 GBfrom $0.055/hRTX 309024 GBfrom $0.241/hRTX 409024 GBfrom $0.308/hLive prices: the cheapest free server in the catalog, updated automatically