Skip to main content
GPUwerk logo

GPUwerk

Dedicated NVIDIA DGX Spark machines, billed per minute

GPUwerk rents whole NVIDIA DGX Spark machines (GB10 Grace Blackwell superchip, 128 GB unified memory), one customer per machine, billed per minute from prepaid credit. It operates its own Sparks in the Czech Republic and rents Sparks owned by partners in Europe, North America and Hong Kong, and also lists an RTX 5090. It is aimed at people weighing a Spark purchase against renting one.

What's good about GPUwerk

  • Rents the DGX Spark by the minute, so a workload can be tested before buying the hardware
  • Live pricing published as a public JSON feed
  • Instances run in the EEA unless another location is chosen
  • Company details, terms and data-processing agreement published on the site

We're actively tracking prices for GPUwerk. Check back soon, or browse other providers with current pricing.

Billing & terms

Pay as you go
Billed per minute from prepaid credit with no fixed term; a stopped instance is billed at a fraction of its running rate
Interruptible
Sparks that can be interrupted, at a lower hourly rate
Reserved terms
Monthly reservations of 1, 3, 6, 12 or 24 months, discounted by term length and billed whether the machine runs or not
Regions
GPUwerk's own Sparks are in Hradec Králové, Czech Republic; partner Sparks are in Czechia, Slovakia, Sweden, the Netherlands, Romania, the United States and Hong Kong. Instances run in the EEA unless another location is chosen
Support
Email support, documentation and setup guides, and a public status page (status.gpuwerk.com)

Watch out for

  • Arm64 only on the Sparks: x86-only images do not run
  • No custom base image, host root, kernel modules or driver changes
  • Docker and systemd are available on some machines only, and the renter cannot choose which
  • Partner-hosted machines carry no SLA, and the partner has physical access
  • Workspaces are not backed up

Features

Dedicated DGX Spark Machines
Each rental is a whole DGX Spark with a 20-core Arm CPU, 128 GB of unified memory and 1 TB of NVMe, with root inside a tenant container rather than on the host
Linked Two-Spark Cluster
A cabled pair of Sparks linked over their 200 Gb/s ConnectX-7 ports, rented together for runtimes that support distributed execution
Per-Minute Billing
Prepaid credit is drawn per minute while a machine runs; a stopped machine keeps its workspace at a reduced rate and terminating deletes it
Interruptible and Reserved Terms
Interruptible Sparks at a lower rate, and monthly reservations of one to 24 months for a fixed machine or cluster
Published Benchmarks
Benchmark pages list model settings and raw vLLM results measured on a single Spark

Company

Published by GPUwerk to identify the company behind this listing.

Legal entity
PRINT IT! SE
Headquarters
🇨🇿 Prague, Czechia
Registration
Commercial Register, Municipal Court in Prague (IČO) 06960090

Getting started

  1. Step 1

    Create an account and add credit

    Sign in to the GPUwerk console and top up prepaid credit by card

  2. Step 2

    Choose hardware and image

    Pick a single Spark, a linked pair or another GPU, an image (DGX OS 7 or Ubuntu 24.04 with CUDA 13.0) and a location, then check live capacity

  3. Step 3

    Connect and run

    Open the browser terminal or connect over SSH with your own key; terminate when finished to end billing

Where this page's facts come from

Profile
Reviewed Nov 18, 2026
Method
How we collect and check prices

Frequently Asked Questions

What GPU types does GPUwerk offer?

Check the pricing table above for GPUwerk's current GPU availability and pricing.

How do I get started with GPUwerk?

Create an account and add credit, Choose hardware and image, Connect and run

What are GPUwerk's main advantages?

GPUwerk's main advantages include: Rents the DGX Spark by the minute, so a workload can be tested before buying the hardware, Live pricing published as a public JSON feed, Instances run in the EEA unless another location is chosen, Company details, terms and data-processing agreement published on the site.

What are GPUwerk's limitations?

GPUwerk's main limitations include: Arm64 only on the Sparks: x86-only images do not run, No custom base image, host root, kernel modules or driver changes, Docker and systemd are available on some machines only, and the renter cannot choose which, Partner-hosted machines carry no SLA, and the partner has physical access, Workspaces are not backed up.

The weekly price report

One email every Monday with the week's notable GPU and LLM API price moves, plus occasional product news. No spam.