Hyperstack GPUs on DeployGPU

The Hyperstack offers on DeployGPU are GPU virtual machines from Hyperstack, launched on DeployGPU's own Hyperstack account and billed from your DeployGPU balance, so you do not need a Hyperstack account. You log in over SSH as ubuntu on port 22. Two things set these VMs apart, and both are covered below: every inbound port is open, and the first SSH attempt can arrive a little before the machine is ready for it.

Offers now

2

From

$0.50/GPU-hr

GPUs per VM

1x, 8x

GPU models

2

Canada, United States · checked 2026-10-13 14:35 UTC

Hyperstack prices and GPUs right now

Each row is one GPU family in stock on Hyperstack, with the number of VM offers, the GPU counts per VM, the countries and the cheapest price per GPU-hour. The prices are the rates you pay on DeployGPU, including any platform fee, refreshed every 15 minutes. Hyperstack's offers are mostly in Canada and the US, with Norway appearing at times.

For Hyperstack's prices next to the rest of the market, see Hyperstack on GpuPerHour.

How a Hyperstack VM differs from the other 3 VM providers

Every inbound port is open: bind to 127.0.0.1

Hyperstack's own default blocks all inbound traffic, SSH included, so DeployGPU creates its Hyperstack VMs with inbound TCP and UDP allowed from any address. That makes SSH work on port 22, and it also means anything you start on 0.0.0.0 is on the public internet within seconds. A Jupyter server without a token, a vLLM endpoint or a database bound to every interface is open to whoever scans the IP.

Bind your own services to 127.0.0.1 and reach them through an SSH tunnel:

ssh -N -L 8000:127.0.0.1:8000 ubuntu@YOUR_VM_IP -p 22

Or turn on ufw and allow only SSH, adding that rule before enabling it so you do not lock yourself out:

sudo ufw allow OpenSSH
sudo ufw enable

The app templates are already safe here: they publish their port only on the VM's 127.0.0.1, and the private link is the way in. The SSH port forwarding guide covers tunnels for everything else.

Connection refused in the first minute after launch

Hyperstack reports a VM as ready once it is active with a public IP, and nothing checks that SSH is answering before DeployGPU shows it as Running. So the first ssh can fail with Connection refused for a short while after the dashboard turns green. Wait a minute and try again. If it still fails after a few minutes, Connection refused on SSH walks through the checks.

One SSH key per VM, and 570 drivers for Blackwell

Hyperstack holds a single SSH key per VM, and the dashboard attaches one key per launch. The deploy page pre-selects the key you used most recently, so check it is the one whose private half is on your computer before you launch. Logging in with a different key fails with Permission denied (publickey).

Hyperstack lists its OS images per GPU type and region. Blackwell machines (B200, B300 and the RTX PRO 6000 server cards) need an NVIDIA driver from the 570 branch or newer, so DeployGPU refuses an image whose name shows an older driver branch with the error image_incompatible, before anything is charged. The provisioning failures guide lists the other launch errors.

Disk, volumes and what you pay

The disk size is fixed per offer, so check the Disk column before you launch: model weights, Docker images and datasets all share it, and the no space left on device fix covers what to do when it fills up. Stopping deletes the VM and its disk. To keep data between sessions, attach a volume created on the Volumes page before you launch. It lives in one region, is billed every hour it exists, and is deleted permanently, with no email, if your balance cannot pay for it.

The hourly price is the rate you are billed, and launching reserves the first hour. When you stop, the unused seconds of the current hour go back to your balance, and a launch that fails is refunded in full. Card top-ups carry a 5% + $0.30 processing fee. The cost calculator works out a run from live prices.