Root GPU Server · for AI training, inference and fine-tuning

Your GPU, your stack. On Swiss hardware.

Full root access to dedicated GPU hardware with Nvidia GPUs, as GPU Standard or GPU Advanced. We deliver the server with Ubuntu LTS and replace defective hardware, you install your drivers, frameworks and models yourself.

Talk to an expert Is this for me? Price Calculator
from CHF 705.– per month plus GPU Up to 2 TB RAM Full root access Swiss location
Jump to
Use Cases Hardware Shared Responsibility Getting Started Pricing FAQ
Use cases

Sound familiar?

“We want our own CUDA, driver and framework stack.”

How Root GPU Server helps

You have the ops expertise and want full root access to install your own CUDA, driver and framework stack, without a managed layer in between.

“Our training and fine-tuning take too long.”

How Root GPU Server helps

Thousands of CUDA cores and dedicated Tensor cores handle compute-intensive model training and fine-tuning, fully under your control.

“We want to run our own LLMs without sending data abroad.”

How Root GPU Server helps

Host large language models yourself on dedicated GPU hardware in Switzerland, with GPU memory from 24 GB up to 96 GB VRAM, operated by an ISO 27001 certified company.

“We render, run simulations or work with CAD.”

How Root GPU Server helps

High memory bandwidth and CUDA core counts suit visualisation, computer-aided design and simulation workloads.

“We need a powerful hypervisor, not a GPU.”

How Root GPU Server helps

Use the same powerful base system without a GPU, with full root access for your own virtualisation platform.

Hardware

Choose your GPU

GPU Standard: H100 NVL · L40s · RTX 4500 ADA

AMD EPYC 7763P with 64 cores, 64 GB RAM (max. 2 TB) and 2x 960 GB NVMe SSD included. Up to 2 GPU slots for the Nvidia H100 NVL, L40s or RTX 4500 ADA.

GPU Advanced: RTX Pro 6000 Blackwell

AMD EPYC 9554 with 64 GB DDR5 RAM (max. 2 TB), no drives included, and up to 4 GPU slots, exclusively for the Nvidia RTX Pro 6000 Blackwell Server Edition with 96 GB VRAM.

Expandable

On request, we equip your server with up to two GPUs (GPU Standard) or up to four GPUs (GPU Advanced), up to 2 TB RAM and expanded NVMe or SATA storage.

Your own operating system

We deliver the server with Ubuntu LTS. Via the IP-KVM, free of charge for the initial setup, you can install other Linux distributions yourself.

Longer commitments

For commitments of 6 months, 1 year, 2 years or more, contact us for an individual quote. You can upgrade to a new card at any time.

Technical details on all GPU models and base systems in our documentation.

Shared responsibility

We take care of the hardware. You run your AI stack.

Root GPU Server is for teams with their own system administration expertise. Everything above the hardware is in your hands.

We handle

The physical server and GPU in our data centre
The installation of Ubuntu LTS
Network connectivity with 1 Gbps flat (Fair Usage)
Repairs under warranty and a replacement within one business day
The Rescue System, for when your server no longer boots

You decide

GPU drivers, CUDA, AI frameworks and your workload
Updates, security patches and firewall
Your backups, for example to our Object Storage
GPU Standard or GPU Advanced, which Nvidia GPU, RAM and storage
Getting started

Your GPU server in four steps

  1. 01

    Choose your server

    GPU Standard or GPU Advanced and your Nvidia GPU, together with our sales team.

  2. 02

    We install it

    We provide the server and GPU with Ubuntu LTS. Other Linux distributions you install yourself via the IP-KVM.

  3. 03

    Root access via SSH

    From here on, you’re in control: install GPU drivers and AI frameworks and manage your workload yourself.

  4. 04

    If hardware fails

    Report it to us and we repair it under warranty. For software problems, use the Rescue System.

Pricing

What does a Root GPU Server cost?

Base server Root GPU Server Standard · AMD EPYC 7763P, plus GPU CHF 705.–
Base server Root GPU Server Advanced · AMD EPYC 9554, plus GPU CHF 1'249.–
Nvidia H100 NVL · 94 GB VRAM (Standard) CHF 3'999.–
Nvidia L40s · 48 GB VRAM (Standard) CHF 999.–
Nvidia RTX 4500 ADA · 24 GB VRAM (Standard) CHF 399.–
Nvidia RTX Pro 6000 Blackwell · 96 GB VRAM (Advanced) CHF 1'749.–

Prices per month, excluding VAT. Plus a one-time setup fee of CHF 700.– for the base server and a one-time fee per GPU card, for example CHF 399.– for an RTX 4500 ADA. Included: hardware rental, initial OS installation, network connectivity and hardware warranty. Additional RAM from CHF 26.50 per 32 GB and month. Full, up-to-date pricing including all RAM and storage options in the Price Calculator. Would rather not manage the server yourself? Have a look at our Managed GPU Server.

Trust

Your models stay in Switzerland

ISO 9001 ISO/IEC 27001 Uptime Institute Tier IV Data centres in Switzerland
«The collaboration with Nine was a real stroke of luck. We finally found a partner in Switzerland who not only understands our specific needs for LLM inference but also has the technical competence to build a perfectly tailored, high-performance, and cost-effective server solution.»
Fabio DuòFounder
Case studies: Peak Privacy View All →
FAQ

Frequently asked questions

GPU Standard or GPU Advanced: which base fits my needs?
GPU Standard runs on an AMD EPYC 7763P with 64 cores, 64 GB RAM (max. 2 TB), 2x 960 GB NVMe SSD included and up to 2 GPU slots, compatible with Nvidia H100 NVL, L40s or RTX 4500 ADA. GPU Advanced runs on an AMD EPYC 9554 with 64 GB DDR5 RAM (max. 2 TB), no drives included and up to 4 GPU slots, exclusively for the Nvidia RTX Pro 6000 Blackwell Server Edition.
Which Nvidia GPU model should I choose?
H100 NVL (94 GB VRAM) suits the most demanding training workloads, L40s (48 GB VRAM) balances performance and cost for inference and rendering, RTX 4500 ADA (24 GB VRAM) fits smaller workloads, and the RTX Pro 6000 Blackwell Server Edition (96 GB VRAM) is our latest GPU generation, exclusively for GPU Advanced. We’re happy to advise you on what fits your workload.
What’s included in the base price, and what isn’t?
We deliver the physical hardware and GPU with Ubuntu LTS pre-installed, and provide the network connection and hardware warranty. Everything above the OS, from GPU drivers and AI frameworks to security patching and ongoing administration, is entirely in your hands. If you’d rather have us manage the server for you, take a look at our Managed GPU Server.
Can I equip my server with more than one GPU?
Yes, on request we can equip your Root GPU Server with up to two GPUs (GPU Standard) or up to four GPUs (GPU Advanced), as well as up to 2 TB RAM and expanded NVMe/SATA storage.
Can I use the base system without a GPU?
Yes. The GPU Standard and GPU Advanced base systems can also be used without a GPU, for example as a powerful hypervisor for virtualisation that you install and configure yourself with full root access.
How do backups work?
For your own backups, you can use our Object Storage (billed separately, not included in the base price). We recommend the duplicity tool for creating and managing backups yourself; a step-by-step guide is available in our documentation.
What happens in the event of a hardware defect?
Parts and repairs are covered by our hardware warranty at no charge. In the event of a serious defect or complete hardware failure, we provide a replacement server with a configuration as close to the original as possible within one business day.
Who is responsible for protecting against brute-force attacks?
For Root GPU Server, that’s entirely your responsibility, for example via your own firewall rules or tools like Fail2ban. Automatic network-level brute-force protection is available with our Managed products.
Can I recover my server myself if something goes wrong?
Yes. Our Rescue System (live Linux) lets you examine and repair your server even when a normal boot is no longer possible, for example after a failed update or a lost password, without our involvement. It requires Linux know-how and does not perform automatic recovery.
Is there an SLA?
No. SLAs with guaranteed response times and service credits aren’t available for Root Server. If that matters for your workload, our Managed GPU Server offers an optional SLA on top.
Is GPU available on Root CloudVM?
No. GPU models are only available for Managed and Root Dedicated Server, not for Root CloudVM.

Technical questions? You’ll find the details in the documentation.

From the docs

Go deeper on docs.nine.ch

All Root Server docs
Prices & product overview →Base systems, GPU cards and add-ons for GPU Standard and GPU Advanced. Offered operating systems →Ubuntu LTS by default, or your own via IP-KVM. Remote Console IP-KVM →Install your own operating system remotely. Hardware defect procedure →What happens when hardware fails. Back up a Root Server →How to set up your own backups.
One provider

One provider, from rack to app.

From colocation to Root and Managed Products, platforms and services: all with us, in our Swiss data centres. The highlighted layer is this page. Grow from one layer to the next without switching providers.

One Swiss teamGrow without switching providers
Services
Databases OpenSearch SLA Monitoring Managed Services Cloudflare
Platform Products
Deploio vCluster Airlock Microgateway WAF
Managed Products
Managed Dedicated Server Managed Virtual Server Managed GPU Server Managed Kubernetes
Root Products
Root Dedicated Server Root CloudVM Root GPU Server
Storage & Backup
Object Storage
Colocation & Housing
Colocation Server Housing
Not quite right?

This might fit better

Rather hand off operations? Managed GPU Server →The same hardware, with server management, monitoring and backups by us. No GPU needed? Root Dedicated Server →Dedicated servers with root access, from AMD EPYC 7313P. Have your own GPU hardware? Colocation →Bring your own hardware to our data centres.

Which GPU fits your workload?

We put the configuration together with you.

Talk to an expert Price Calculator

Talk to our experts

Do you have questions about our products? No problem. Contact our sales department or go directly to your Cockpit to see all the possibilities at Nine.

Contact Sales Go to Cockpit

Subscribe to our newsletter for the latest updates: