Hello,

If you’ve tried running AI inference on a CPU-only VPS, you know how it goes: the job finishes eventually, but you’ll have time to question your life decisions while the progress bar crawls.

GPU VPS fixes that. Launching today, it’s our first product to put a dedicated NVIDIA GPU behind the flat monthly pricing and fast setup you already get from a Contabo Performance VPS.

What you get

The usual VPS Add-Ons come along too, and the 1-Click App library with AI apps. It’s one flat fee with unlimited traffic (fair usage policy applies).

What it’s for

For AI and ML teams, this is home base. Inference, fine-tuning, and LoRA training on your own models all sit on a single card, without token metering or a full-server commitment.

Research and engineering teams get a proper single-node machine for CFD, molecular dynamics, FEM, and data science: work that’s painful on CPU but doesn’t justify a multi-node cluster.

And for rendering and FX, one dedicated card takes 3D, motion design, and compositing off your local workstation, and handles the heavy usage periods a single machine can’t keep up with.

The card we chose, and why

The 96 GB of VRAM is what earns this card its place: it holds a 70-billion-parameter model on one card without splitting it across instances. Renting it monthly gets you that hardware without tying up your money or watching your card lose value when the next architecture is released.

The full breakdown is on the blog: read it here.

One more thing. We’ve open-sourced our product documentation, built with Hugo and live at docs.contabo.com, GPU VPS included. It’s developed in the open on GitHub, so you can open issues and send PRs as it grows. We’re putting it out early on purpose, because we’d value your feedback. Let us know what you think!

Your Contabo Team