GPU VPS

Dedicated GPU VPS for AI Training & Inference

Name: GPU VPS
Brand: Contabo
Availability: InStock

Run your own LLMs and AI models on a dedicated NVIDIA RTX PRO 6000 with 96 GB of VRAM. Fine-tune a model, serve inference, or build a RAG pipeline minutes after checkout. CUDA is pre-installed and the whole GPU is yours, with no sharing and no GPU slicing.

Dedicated GPU resources. Monthly billing available. No hidden costs. NVMe storage.

GPU VPS

€999.00

799

20 / month

for the first 24 months

Get Started

Dedicated NVIDIA RTX PRO 6000

CUDA-compatible・NVMe storage

Unlimited traffic*

96 GB

GDDR7 VRAM

24,064

CUDA CORES

120

FP32 TFLOPS

1,597

GB/s Bandwidth

Step 1: Choose your Location

Pick your GPU VPS region (Europe or US Central).

Step 2: Finalize your order

Your instance deploys automatically on Ubuntu 24.04 with CUDA drivers, the CUDA toolkit, and nvidia-container-toolkit pre-integrated.

Step 3: Connect and start working

SSH in and go. With the basic setup done for you, you can launch any CUDA-compatible framework the moment you connect.

background-image_gpu-vps-why-run-ai-training_alt

100% Dedicated GPU Resources

The full RTX PRO 6000 is mapped to a Performance VPS with no vGPU slicing. All 96 GB stays yours, so throughput holds from the first training step to the last.

CUDA-Compatible Infrastructure

Every GPU VPS comes with Ubuntu 24.04 with CUDA already installed. Pull a PyTorch or vLLM container and get straight to work.

Built for Large Models

Blackwell architecture with native FP4 and 96 GB of VRAM runs models a consumer card cannot hold. A 70B-class model serves at FP8 on a single card.

Transparent Pricing

One flat monthly rate (annual billing available) with unlimited traffic (fair usage policy applies). No per-second or per-token metering.

Your Infrastructure on Your Terms

Root access and standard Linux tooling mean total control. Keep your weights, prompts and user data in a European or US data center you choose.

24/7 Support

Award-winning support from both human specialists and high-quality AI-assisted responses. Quick answers when a training run or an endpoint needs them.

RTX PRO 6000

120

Blackwell Server Edition · 96 GB GDDR7 ECC · 24,064 CUDA cores

L40S

91.6

Ada Lovelace · 48 GB GDDR6 · 18,176 CUDA cores

H100

Hopper · 80 GB HBM3 · 16,896 CUDA cores · built for multi-node training

H200

Hopper · 141 GB HBM3e · 16,896 CUDA cores · built for multi-node training

A100

19.5

Ampere · 80 GB HBM2e · 6,912 CUDA cores · previous generation

FP32 compute (TFLOPS) - higher is better

LLM Inference with Your Own Model

Serve your own weights and keep every prompt on infrastructure you control. A 70B-class model fits at FP8, with room left for context.

Fine-Tuning and LoRA Training

LoRA y Fine-Tuning de parámetros completos en una sola tarjeta. Los 96 GB de VRAM le permiten cargar modelos base que una GPU de consumo de 24 GB no puede cargar.

RAG Pipelines at Scale

Run the embedder, vector store and generation model on one GPU. No network switching between them, and your data stays in your environment.

Dev & Test Environments for AI Teams

One always-on GPU box for the whole team, pre-configured with CUDA. No cold starts and no re-provisioning between sessions.

OpenClaw

n8n

Hermes Agent

Ollama

Dokploy

Paperclip

Coolify

cPanel

Plesk

Docker

Webmin

Proxmox

Deploy Your Perfect Image

With Custom Images, you can instantly deploy your .iso or .qcow2 image via web UI or API.

See more →

What is a GPU VPS?

A GPU VPS is a Virtual Private Server with a dedicated NVIDIA GPU attached via PCI passthrough. Unlike shared GPU instances, the full GPU, including VRAM, CUDA cores, and memory bandwidth, is reserved for your workload alone.

Which GPU does Contabo GPU VPS use?

Every GPU VPS runs the NVIDIA RTX PRO 6000 Blackwell Server Edition, with 96 GB of GDDR7 ECC VRAM, 24,064 CUDA cores, 120 TFLOPS of FP32 compute, and 1,597 GB/s of memory bandwidth.

Is the GPU dedicated or shared on Contabo GPU VPS?

The GPU is dedicated. The full RTX PRO 6000 is allocated to your VPS through 1:1 PCI passthrough, with no vGPU slicing.

Is this GPU VPS good for training, or only for inference?

Both. A dedicated RTX PRO 6000 with 96 GB of VRAM and 5th-generation Tensor Cores handles training, fine-tuning and inference. Fine-tuning and adapting existing models is its strongest use, which covers most real-world training work.

Can I run LLM inference and fine-tuning on the same GPU VPS?

Yes. The usual workflow is to fine-tune your model on the GPU VPS, then serve it for inference on the same dedicated RTX PRO 6000. Both are single-GPU workloads, and 96 GB of VRAM gives room to move from one stage to the next.

Can I run multi-GPU or multi-node AI training on a Contabo GPU VPS?

A single GPU VPS is built for fine-tuning and adapting existing models on one dedicated RTX PRO 6000 with 96 GB of VRAM. For large-scale training across multiple GPUs or nodes, our sister company vshosting offers dedicated multi-GPU servers with NVLink, H200 and B300 configurations.

Can a 70B parameter model actually run on this GPU VPS?

Yes, at FP8 precision. A 70B-class model's weights occupy roughly 70 GB at FP8, one byte per parameter, leaving headroom within 96 GB of VRAM for context and inference overhead.

Which AI models are confirmed to run on GPU VPS?

Stable Diffusion, Qwen3-14B, Nemotron-Nano-3-30B, Devstral-2-123B, Qwen-Image-2512, and Coqui TTS are all confirmed. Any other CUDA-compatible model that fits within 96 GB of VRAM will run.

Is CUDA pre-installed?

Yes. Every GPU VPS ships on Ubuntu 24.04 with CUDA and nvidia-container-toolkit already installed.

What operating systems are supported?

Ubuntu 24.04 LTS with CUDA pre-installed is the only supported image for GPU VPS in the order process. Windows can be installed later using the Customer Control Panel if desired - in this case, you will need to configure CUDA yourself.

Where are GPU VPS data centers located?

Europe and US Central are the only available regions at launch. Starting in two locations is a deliberate first step. GPU capacity is harder to plan than standard VPS, so we are keeping the launch focused to protect availability and performance, with more regions planned.

Learn More About GPU VPS

View all articles →

Tutorial

How to Run LLM Inference on a GPU VPS

Set up a private LLM endpoint, step by step.

Contabo Blog

blog-ref-image_gpu-vps-for-ai-training-and-inference_alt

Tutorial

GPU VPS for AI Training and Inference: A Practical Guide

How to size a GPU VPS for training vs. inference.

Contabo Blog

Knowledge Base

GPU VPS Documentation

Contabo guides and API reference, GPU VPS included.

Contabo Help Desk

The price displayed is the effective monthly rate for a 24-month subscription including applicable taxes.

Minimum initial contract length: 1 month | Minimum following contract length: Equal to the initial contract length | Minimum cancellation notice: None (pre-paid) or 4 weeks (post-paid).

Incoming Traffic: Unlimited and unmetered. No extra charges apply. | Outgoing Traffic: Unlimited - fair usage policy applies based on average server workloads. To maintain fair network performance for all customers, Contabo reserves the right to throttle servers with exceptionally high or disruptive usage patterns.

Applications and Add-Ons: Contabo provides full technical assistance for the Contabo VPS itself - including the network, hardware, and initial setup of any 1-click Add-Ons. Any additional open source software included in the applications is owned by the respective provider - maintenance, upgrading, and troubleshooting are within the end-users responsibility.

Dedicated GPU VPS for AI Training & Inference

Run your own LLMs and AI models on a dedicated NVIDIA RTX PRO 6000 with 96 GB of VRAM. Fine-tune a model, serve inference, or build a RAG pipeline minutes after checkout. CUDA is pre-installed and the whole GPU is yours, with no sharing and no GPU slicing.

Dedicated GPU resources. Monthly billing available. No hidden costs. NVMe storage.

DEDICATED NVIDIA RTX 6000 PRO
FOR EVERY AI WORKLOAD

Get started in minutes

Step 1: Choose your Location

Step 2: Finalize your order

Step 3: Connect and start working

Why Run AI Training and Inference Workloads on a Contabo GPU VPS?

How the RTX 6000 Pro Measures Up

GPU VPS Use Cases for AI Teams

Customize Your GPU VPS Your Way with Apps and Deployment Options

Looking for inspiration? Browse AI
models and apps on Hugging Face.

FAQs

Learn More About GPU VPS