DGX Spark/How to rent
Getting started

How to rent a DGX Spark

By Samuel Seidel · Published September 14, 2026 · Updated October 24, 2026 · 8 min read

A DGX Spark rents for $0.79/hour, billed per minute from prepaid credit, and your first $20 top-up comes with $30 in credit. Sign in, top up, deploy, then open the browser terminal, which needs no SSH key, or connect as root over SSH with your own key. The six steps below follow the console in order.

At a glance

ItemDetail
HardwareDGX Spark GB10 (ARM64), 128 GB unified memory, 1 TB NVMe shared with the system and model cache
Price$0.79/hour, billed per minute, prepaid, excluding VAT per the terms. A two-node cluster is $1.79/hour. See pricing.
First top-upPay $20, get $30 in credit. The minimum top-up is $20.
LocationEurope (EEA) by default: GPUwerk's own Sparks in Czechia and partner Sparks in other EEA countries. Partner Sparks in the US and Hong Kong only if you choose them; see locations and Partner Sparks below. Details: GDPR and the EU AI Act on dedicated hardware.
TenancyDedicated, one tenant per machine
AccessBrowser terminal, or root SSH to your own container (root inside it, not on the host). CUDA 13.0 toolkit preinstalled. systemd and Docker inside the rental on some machines only; the others run a plain container without them, and you can't choose which (details).
StoppingHeld at 75% of the rate, $0.59/hour, with your workspace kept. Terminate deletes the workspace.
Other providersOther DGX Spark rental offers are listed, with sources, on the cloud providers page.

Checked against GPUwerk's pricing page and terms on October 24, 2026.

How much does it cost to rent a DGX Spark?

One DGX Spark costs $0.79 an hour, billed per minute, so a minute costs about 1.3 cents. Prices exclude VAT. Here is what that comes to if the Spark runs the whole time:

Running timeCost at $0.79/h
1 hour$0.79
8 hours$6.32
1 day (24 hours)$18.96
1 week (168 hours)$132.72
20 hours a month$15.80 a month
80 hours a month$63.20 a month
160 hours a month$126.40 a month
Always on (730 hours)$576.70 a month, or less on a monthly reservation

A stopped Spark is billed at 75% of the rate, $0.59 an hour, so terminate it when you no longer need the workspace. A two-node cluster is $1.79 an hour. All rates are on the pricing page.

Can I rent a DGX Spark without a sales call?

Yes. You can rent a DGX Spark self-serve. You sign in with Google or GitHub, top up by card through Stripe, and deploy from the console. Nobody has to approve you and there is no call. If a Spark is free, you can be in the browser terminal, or connected over SSH, within minutes. If every Spark is in use, the console offers the waitlist.

Can I rent for one day or by the minute?

Yes. There is no minimum term for on-demand instances. Billing is per minute from prepaid credit, so a short test costs only the minutes it runs and a full day costs $18.96. To deploy, your balance has to cover one hour of running time. Terminate the instance when you are done and billing stops.

Create an account

Sign in at console.gpuwerk.com with Google or GitHub. Signing in doesn't commit you to anything or start any billing.

Add prepaid credit

Rental is prepaid, not invoiced after the fact, and you pay through Stripe, which handles the card details on its own page. The minimum top-up is $20. Get $10 in bonus credit on your first qualifying top-up of $20 or more. Credit is a balance, not a fixed-duration booking. It is spent down as instances run, at $0.79/hour for a single Spark or $1.79/hour for the two-node, 256 GB cluster. Prices exclude VAT, which is added according to your billing details. Adding credit does not reserve a machine.

Add an SSH key, if you want SSH access

You don't need a key to start a Spark or to use the browser terminal. Add one if you want to connect from your own computer over SSH. The console installs your saved keys when an instance is created. If you add a key after deploying, use Reset SSH access on the instance page to install it; nothing else on the instance changes. SSH accepts keys only, so there is no password fallback. If you don't have a key pair yet, the SSH setup guide covers generating an ed25519 key and loading it into your agent. Paste only the public key into the console and keep the private key on your own machine.

Start a Spark

The deploy screen has three parts: hardware, image, then access and running cost. Pick a single Spark or the cluster. For the image, a base image (DGX OS 7 or Ubuntu 24.04 LTS) gives you a CUDA environment to install and run whatever you like. Model images start a model server for you, and the console lists these presets:

There are also tool images: ComfyUI, JupyterLab and VS Code in the browser. A model image also gets a public HTTPS endpoint on its hostname, which needs an API key with the inference scope; set up access controls before you send it sensitive inputs. If no Spark is free, the console offers the waitlist. Once the instance reaches ready status, open the terminal from the instance page, or copy the SSH command from the instance details if you added a key.

Connect

The browser terminal needs no key and no SSH client: it opens a shell on your instance from the console. SSH drops you into your own container as root, with CUDA already installed (and Docker, on some machines), so you can run a serving stack you choose or use whatever the image already started. The SSH guide also covers a passwordless alias, so the connection is one word to type.

Stop or terminate when you're done

These are two different actions with two different costs. Stopping does not end billing.

Copy anything you need off the instance before you terminate it.

What happens when credit runs out

When your credit reaches zero, or you hit a monthly budget you set, the instance stops and its machine is released. The workspace stays on that machine for up to 48 hours. Top up (or raise the budget) and press start in that time to pick up where you left off, unless someone else has rented the machine, which deletes it sooner. After 48 hours it is deleted for good. Low-balance emails warn you first, and automatic top-up lowers the risk, but a payment can fail.

GPUwerk keeps no copy of your workspace: no backup and no recovery copy. If a drive fails, the workspace on it is lost, so keep your own backups.

What you work with once it is running

A rented Spark is one dedicated machine with a shell, a model endpoint if you want one, and an API. Nothing here needs a sales call or a custom setup.

Partner Sparks

The console offers one Europe (EEA) location for GPUwerk's own and the partners' EEA Sparks, and each partner country outside the EEA as its own location. The machine is in the EEA unless you choose a location outside it, or Anywhere, in the Location menu, and the instance page shows its country once it is placed; there is no SLA, and the partner has physical access to the machine. You may use any data on it; choosing a location that suits your data is up to you. That is the one exception to EU-Central on this page.

Plan the session before you start it

At $0.79/hour, eight hours of running time is $6.32. Decide roughly how long your test will take and load enough credit to cover it comfortably, because running out mid-session stops the instance and releases its machine, and its workspace is deleted once someone else rents that machine, or after 48 hours at the latest. The first-evaluation guide plans a specific model test end to end, including the commands to run once you're connected.

Renting against buying

Whether to rent at all or buy the hardware is a separate question from the mechanics on this page. The price page lays out the EU purchase range next to the rental rate, and the rent-vs-buy calculator works out the break-even point for your own usage pattern. If you would rather own one and have GPUwerk house it and rent it out, see the GB10 store.

FAQ

How much does it cost to rent a DGX Spark?

$0.79 an hour for one Spark, billed per minute, excluding VAT. That is $18.96 for a full day, $132.72 for a week, and $15.80, $63.20 or $126.40 a month at 20, 80 or 160 hours. Always on, it is $576.70 a month, or less on a monthly reservation.

Can I rent a DGX Spark without a sales call?

Yes. Renting is self-serve: sign in with Google or GitHub, top up by card through Stripe, and deploy from the console. If a Spark is free, you can be in the browser terminal or on SSH within minutes.

Can I rent for one day or by the minute?

Yes. There is no minimum term for on-demand instances, and billing is per minute from prepaid credit. A short test costs only the minutes it runs, and a full day $18.96. To deploy, your balance has to cover one hour of running time.

How do I sign in to the console?

With Google or GitHub sign-in at console.gpuwerk.com. Signing in starts no billing.

How much do I need to add to start?

Top-ups are prepaid credit against the $0.79/hour on-demand rate, billed per minute. The minimum top-up is $20. Get $10 in bonus credit on your first qualifying top-up of $20 or more.

What's the difference between stopping and terminating?

Stopping keeps your workspace on the same node and bills 75% of the running rate ($0.59/hour) to keep the Spark held for you. Terminating ends billing entirely and deletes the workspace for good, which is why it asks you to type the instance name to confirm.

Do I need my own inference server, or is one preinstalled?

Both are possible. You get root SSH into your own container with the CUDA 13.0 toolkit preinstalled (and, on some machines, Docker), so you can install anything yourself, and some images come preconfigured with a model already served, such as the Qwen3 Coder 30B image used in the first-evaluation guide.

Where is the hardware located?

GPUwerk's own Sparks are in Hradec Králové, Czechia. Partner Sparks are in other countries: on 10 November 2026 in Czechia, Slovakia, Sweden, the Netherlands, Romania, the United States and Hong Kong. The console's Europe (EEA) location covers GPUwerk's own and the partners' EEA machines; a machine outside the EEA is only used if you choose its country, or Anywhere, in the Location menu. Partner machines have no SLA. You may use any data on it; choosing a location that suits your data is up to you.

Is there a waitlist?

Not for the Spark when a machine is free: you deploy and it is yours. When every Spark is in use, the console offers a waitlist, and people in line are emailed one at a time, oldest first, with no machine held for anyone; whoever deploys first gets it. The x86 GPUs (RTX 5090, RTX 6000 Ada, L40S, RTX PRO 6000 and H100) work the same way.

Related reading: current availability, full specs, what it costs to buy instead, and where to buy one in Europe if renting isn't the answer for your workload.

First top-up: pay $20, get $30 in credit

Rent your first Spark.

A dedicated DGX Spark for $0.79/h, billed by the minute. Create an account, add credit, and check live availability.

Create account See full pricing