Frequently Asked Questions

GamePC rents you a cloud machine with one dedicated NVIDIA GPU. The machine is a Linux container with PyTorch and CUDA preinstalled. You open it in your browser through Studio (JupyterLab) or connect over SSH, and you pay for the time it runs, by the second. There is no hardware to buy or maintain.

No. GamePC machines have no display, no graphical desktop and no game streaming — you can't see or control a screen on them. They are built for compute work: AI and machine learning, notebooks, data processing and rendering or batch jobs run from the command line.

A computer with a current web browser, and a GamePC account with at least $1.00 of credit (top-ups start at $1.15). An SSH client is optional — it is built into macOS and Linux, and OpenSSH or PuTTY are available elsewhere.

Machines run on GPU hosts from the Vast.ai marketplace. When you deploy, we rent a host that has your tier's GPU free, wherever it is. You can't choose a region.

GamePC uses prepaid credit. You add credit from $1.15, then deploy a machine on one of four GPU tiers, each with an hourly rate. A running machine is charged by the second; when you stop it, the charges stop. There are no subscriptions, contracts or setup fees.

Only while your machine is in the Running state. Stopped machines are not charged, and a deploy that doesn't start within 10 minutes is cancelled and not charged. You can see every charge on the Billing page in your dashboard.

Visa and Mastercard credit and debit cards. Payments are processed by our card payment partner, and your bank may ask you to confirm the payment with 3-D Secure.

Unused credit can be refunded within 14 days of purchase. If a running machine stops working because of a fault on our platform or on its GPU host, email support@gamepc.io with the machine ID and we credit the time you lost. Refund requests go to support@gamepc.io too.

If the balance reaches zero while a machine is running, the machine is stopped automatically. Its disk is kept for 24 hours, so you can add credit and start it again; after that it is deleted. If a machine is running and your balance is under $2.00, we email you a warning — we check once a day, so a fast-running balance can reach zero before that email arrives.

Four tiers: Starter (RTX 3060, 12 GB VRAM) at $0.40/hr, Standard (RTX 3090, 24 GB VRAM) at $0.60/hr, Pro (RTX 4090, 24 GB VRAM) at $0.90/hr and Power (RTX A6000, 48 GB VRAM) at $1.20/hr. Each machine gets one whole GPU and a 100 GB disk. A tier can be out of stock or temporarily unavailable; the pricing and deploy pages show which.

Linux. Each machine is a container started from the official PyTorch image, with PyTorch, the CUDA libraries it uses, JupyterLab and SSH. You are root inside the container, so you can install more with pip or apt-get. The CUDA compiler (nvcc) is not included; install the CUDA toolkit yourself if you need to compile CUDA code.

Usually within a few minutes. It depends on the GPU host the machine is placed on. If a machine has not started within 10 minutes we cancel the deploy and you are not charged. Once it is running, the Open Studio button and the SSH command appear on the machine's page.

For 24 hours. Stopping keeps the disk, and starting the machine again within 24 hours brings your files back. A restart needs the same host's GPU to be free, so it can be delayed if someone else has rented it in the meantime. After 24 hours the machine and its disk are deleted, and deleting a machine yourself deletes it straight away — download anything you want to keep.

Your machine runs in its own container, and its GPU is not shared with anyone else while it runs. The hosts belong to third-party providers on the Vast.ai marketplace, so don't put data on a machine that you are not allowed to process on third-party infrastructure.

Two ways. Click Open Studio on the machine's page to open JupyterLab in your browser — notebooks, a terminal and a file browser. Or use the SSH command shown on the same page from your own terminal.

Each machine serves Studio with its own self-signed certificate, which browsers don't recognise. The connection is still encrypted. Choose Advanced, then Proceed, to continue.

SSH logs in with a key, not a password. Open a terminal in Studio and add your public key with printf '\n%s\n' "YOUR_PUBLIC_KEY" >> ~/.ssh/authorized_keys (on your own computer it is usually ~/.ssh/id_ed25519.pub). Then run the SSH command from the machine's page. You log in as root.

In Studio, use the upload button in the file browser, and right-click a file to download it. Over SSH, use scp or rsync with the same host and port as the SSH command.

Open a terminal in Studio and run nvidia-smi — your GPU is listed. In a notebook, import torch and run torch.cuda.is_available(); it returns True.

Anything that runs on Linux from a notebook or a terminal and uses an NVIDIA GPU through CUDA: training and fine-tuning PyTorch models, inference with libraries such as Hugging Face Transformers and Diffusers, data processing, and command-line rendering such as Blender in background mode. Programs that need a graphical desktop can't be used.

Start it with nohup, for example nohup python train.py > train.log 2>&1 &, and it keeps running after your SSH or Studio session closes. Charges continue while the machine runs, so stop it when the job is done.

Check that you have at least $1.00 of credit. If the tier is out of stock, try another tier or check back later. A deploy that hasn't started within 10 minutes is cancelled and not charged, and you can deploy again straight away. If it keeps failing, contact support@gamepc.io with the machine ID.

Make sure the machine shows Running in your dashboard. If the Open Studio button isn't there yet, the machine is still getting its address — refresh in a minute. If the browser shows a certificate warning, choose Advanced, then Proceed.

The machine doesn't have your public key yet. Open a terminal in Studio, add your public key with printf '\n%s\n' "YOUR_PUBLIC_KEY" >> ~/.ssh/authorized_keys, and connect again as root with the command from the machine's page.

Run nvidia-smi in a Studio terminal. If it shows no GPU, delete the machine and deploy a new one, and tell support@gamepc.io the machine ID so we can check the host.