A whole GPU for you, not a slice of someone else’s.
A whole RTX A4000 inside your machine, or a remote session with CPU and memory. In Spain, over SSH with your key, paid from the balance you already have.
Monthly fee, paid from your balance
RTX A4000
16 GB of VRAM, all yours
- Architecture
- NVIDIA Ampere
- Cores
- 6,144 CUDA · 192 Tensor
- Card memory
- GDDR6 with ECC · 448 GB/s
- System
- Debian 13 with NVIDIA driver
Remote sessionNo GPU
What it is for
- Development and editors over SSH
- Agents and background jobs
- Anything that does not touch the GPU
Loading the catalogue…
You choose and configure
Cards, disk and bandwidth. The price recalculates as you move the controls.
You sign in
The same Nextbit account. You give us a name for the machine and your public SSH key.
It is charged to your balance
The first month comes out of the balance you already have. No credit card.
The machine builds itself
In one to three minutes. You watch the progress step by step and end up with the SSH command ready to copy.
It builds itself. No tickets, no calls.
GPU
Root over SSH. With your key, no passwords.
If anything fails while building it, you are not charged. It is delivered checked, or not at all.
What one card costs per month, by commitment.
More cards in one machine, less for each.
GPU, CPU, memory and disk, in sight as soon as you open it.
Share it. You decide what each person can do.
Get in by its name. One command sets up your ~/.ssh/config.
$ ssh my-machine
root@my-machine:~#
More disk without reinstalling. You see the quote before applying it.
8B 8 bits: fits
Takes ≈ 9.6 GB of 16 GB. 6.4 GB to spare.
Estimate: weights = parameters × bytes (2 in fp16, 1 in 8-bit, 0.5 in 4-bit) plus 20% for context.
The whole physical card. Not sliced, not shared in turns: its memory and its cores are yours.
A machine, not a container. Whatever you install stays until you terminate it.
- Install whatever you want
- NVIDIA driver already installed
- You add CUDA, Docker and PyTorch
Isolated from the rest. No other customer can see its traffic.
Our own hardware in Spain. No US parent company.
- Spanish company
- Our own facilities
- Data in the EU
- Out of reach of the CLOUD Act
If you only want to call a model, you do not need a machine: the inference API is cheaper and there is nothing to set up. See the inference API
Is the GPU really dedicated?
Yes. You get the whole physical card, exclusively, for as long as the contract lasts. It is not shared between customers or sliced into turns: its memory and its cores are yours, and nobody else runs anything on it.
Why can’t I choose RAM and cores on the GPU line?
They are tied to the number of cards, and the ratio is set so each card carries its fair share of memory and cores. If you could order one card with all the available memory, the rest would be left unusable. On remote sessions, where there is no card, memory and cores are chosen freely.
How do I pay?
With the balance in your Nextbit account, the same one you use for inference. The first month is charged when you order and then once a month after that. No card needed.
How do I get into the machine?
Over SSH with your public key, which you paste when ordering. We do not use passwords. From the panel you can generate a command that leaves your ~/.ssh/config ready, so you get in by typing just the machine name.
Can I upgrade it later?
Yes, always upwards while the term lasts. You see the quote before anything is applied and only the difference for the remaining days of the period is charged.
Do I pay while it is off?
Yes. The machine and its disk stay reserved for you whether it is running or not: turning it off frees no hardware and gives it to nobody else. To stop paying you have to terminate it.
What happens if I terminate it?
The disk is destroyed. There is no bin and no going back, so copy out whatever you want to keep first. The machine disappears and its cards go back to the pool.
Can I share it with my team?
Yes. From the panel you give access to whoever you want and choose what each person can do: see metrics, restart, add SSH keys, resize it. It stays billed to whoever ordered it.
Where is the hardware?
In our own facilities in Spain. Each machine is isolated from the others: its traffic is not visible to any other customer.
What comes installed?
The GPU one, Debian 13 with the NVIDIA driver already installed. You install CUDA, Docker and PyTorch yourself, in the version you need. The remote session comes with Claude Code and tmux.
Do you back up the disk?
No. The disk is yours and so are its backups: keep anything you cannot afford to lose outside the machine.