Dedicated compute · EU-Central

Rent a dedicated DGX Spark in the EU.

Billed per minute, applicable tax extra. Rent one Spark or a linked pair, or reserve by the month and save up to 35%.

Deploy now Reserve monthly

First top-up: pay $10, get $20 in credit.

Already have an account? Sign in.

$0.79Per hour for one dedicated Spark, billed per minute
128GBUnified memory, shared by the CPU and GPU, with 273 GB/s of bandwidth
1PFLOPFP4 AI compute, sparse, from the NVIDIA GB10 Grace Blackwell Superchip
256GBTwo linked Sparks as one cluster, $1.79 per hour

One dedicated Spark

NVIDIA GB10, by the minute.

Hourly rate
$0.79, billed per minute
8 hours running
$6.32 before tax
Included
20-core Arm CPU, 1 TB NVMe workspace
Deploy now Stopping retains the machine at 75% of the hourly rate. Terminating ends billing and deletes the workspace. See billing and data retention.

Choose how you rent

A short evaluation or a longer commitment.

$0.79/ hour

Pay as you go

Start and stop whenever you like. Billed per minute from prepaid credit, with no fixed term.

One Spark
$0.79 / hour
Two linked Sparks
$1.79 / hour
Stopped
75% of the running rate
See pricing
35%off over 24 months

Reserve a term

Keep the same Spark or cluster for months, billed monthly. The longer the term, the lower the rate.

1 month
On-demand rate
3 months
10% off
6 months
15% off
12 months
25% off
24 months
35% off
You pay for the whole term, whether the machine is running or stopped. Reserve

Beyond the rental

Everything around the GB10.

$0.79/ hour

Rent

A dedicated DGX Spark by the minute, or a linked pair at $1.79 per hour. One RTX 5090 is rentable now at $0.89 per hour; the L40S is on a waitlist.

See GPUs
€5,330excl. VAT

Buy

Every maker's GB10, shipped from Czechia. The Lenovo ThinkStation PGX is on sale from €5,330 excluding VAT and shipping; the other makers' machines are priced from Czech suppliers' stock whenever a supplier can deliver them.

Open store
70%of renter charges

Rent out your Spark

Keep your Spark at your home or office. GPUwerk rents it to its customers and you get 70% of what they are charged for your machine's hours, paid monthly in euros.

How it works
4optional steps

Services

An opportunity session, a workshop and implementation support, then run it on a rented Spark, your own GB10 or a hosted one. The founder runs the first sessions remotely.

Services

Inspect the evidence

Benchmark the workload. Keep the configuration.

Our benchmark pages include model settings and raw results. Compare latency and throughput at the concurrency you need, then run your own evaluation.

Benchmarks

128 GB is unified memory

The CPU and GPU share it. Model weights, runtime, and context all consume memory. Fitting a model does not guarantee the response time your users need.

Read the hardware details →

GPUwerk allocates one tenant per physical machine. Your shell runs in a container; this is not host-level root access.

Quickstart

From account to first run.

  1. Sign in and add credit

    Top up prepaid credit by card. It is billed per minute while a machine runs; applicable tax is extra.

  2. Choose and configure

    Choose your hardware and image, then check live capacity when you deploy. GPUwerk configures the machine, so no SSH key is needed; add your own if you want SSH access. Prepaid credit does not reserve a machine.

  3. Run, measure, and export

    Open the browser terminal in the console, or connect over SSH with your own key. Save your results externally and terminate when finished.

Open the quickstart · Optional: connect over SSH · Understand running and stopped charges

Start with a real task

What do you want to test?

Choose the guide closest to your project. Model fit and performance depend on your configuration.

Coding workflows

Run a coding model

Connect your tools to an open-model endpoint. Measure response speed on your own tasks.

Read the coding-agent guide →
Private assistants

Try a self-hosted chat stack

Explore Open WebUI with your model. Start with public or synthetic test data.

Read the setup guide →
Hardware evaluation

Decide whether to buy a Spark

Check memory needs, Arm compatibility, and your expected traffic before committing to hardware.

Read the buyer’s guide →

Prefer help?

Scope the evaluation together first.

Describe your model, expected context length and what a good result looks like. GPUwerk proposes an evaluation plan on public or synthetic data before you commit to anything.

Get advice

Prefer a structured start? See the first engagement options. Please leave out confidential data.

Before you deploy.

What does a Spark rental include?

A dedicated DGX Spark with 128 GB unified memory, 1 TB of workspace storage, and access to your tenant container from the browser terminal in the console, or over SSH with your own key. Select an image and check current capacity in the console. You manage your software and backups: workspaces are not backed up, so keep your own copy of anything you need.

Will my model run on a DGX Spark?

It depends on the model, quantization, context length, and software support for Arm and GB10. The 128 GB memory is shared by the CPU and GPU; some is needed by the system and runtime. Use our benchmark reports as a starting point and test your own workload.

What happens when I stop or terminate an instance?

Stopping keeps the Spark reserved and bills 75% of its running rate. Terminating ends compute billing and deletes the workspace. Reaching zero credit or the configured budget stops the instance and releases the machine; the workspace is saved for 7 days, then deleted. Export anything you need before that happens.

Where does my workload run, and who operates it?

The compute hardware is in EU-Central and GPUwerk is operated by PRINT IT! SE. Dedicated tenancy does not remove infrastructure administrator access. Review our data processing agreement and security disclosures before uploading sensitive data.

Can I test before buying hardware?

Yes. Rent a Spark to evaluate your workload first. If you then want to own one, GPUwerk's GB10 store sells the same GB10, 128 GB machine from eight makers, and GPUwerk can house yours and rent it out for you, or you can ask about a machine for your own site.

Data processing agreement · Security and privacy disclosures · Service status

Want to own the hardware? Use your rental results to decide: buy a GB10 machine, shipped to you or hosted and rented out by GPUwerk if you choose, or ask about installation and support.

New · GPUwerk Platform

Own a DGX Spark? Rent it out through GPUwerk.

70%

Your share of renter charges

Hosting is open to every Spark owner. Your Spark stays at your home or office, GPUwerk rents it to its customers, and you get 70% of what they're charged for your machine's hours, paid monthly in euros. Open to owners in any country not under EU sanctions. How many hours get rented depends on demand GPUwerk cannot promise.

Start hosting How the earnings work · About the programme · No Spark yet? Buy a GB10 machine and have GPUwerk host it

First top-up offer

Top up $10. Get $20 in compute credit.

Top up $10 and we match it: $20 in credit for your first top-up.

Once per account · $10 minimum qualifying top-up · applicable tax extra. The match applies to compute credit. $20 runs one Spark for about 25 hours.

Top up $10, get $20