NVIDIA RTX 4090 servers in Switzerland

Rent a dedicated bare-metal NVIDIA GPU server hosted in Switzerland, with exclusive GPU, CPU, memory and storage resources. Compare current hourly prices and choose hardware for your workload before signing in.

Standard Linux rentals provide SSH access. You manage the applications and models on your server.

NVIDIA RTX 4090

An Ada Lovelace option for ComfyUI, LLM inference and PyTorch experiments within a 24 GB memory budget. Model weights, context, image resolution and batch size all contribute to memory use.

VRAM
24 GB
CPU
16 cores
RAM
32 GB
Storage
1 TB NVMe

Ada Lovelace · Bare-metal rental: 0.40 CHF/h

Availability: Available.

Rent this GPU · NVIDIA specifications

Choose a GPU for your workload

Choose enough GPU memory for your model weights, working memory, context and batch size. Use the guides below to select your software and get started with your workload.

  • ComfyUI and image generation: check your checkpoint, custom nodes and image resolution before choosing GPU memory.
  • LLM inference with Ollama: choose a model and quantization that fit, allowing memory for the context and concurrent requests.
  • PyTorch development: check CUDA, framework support and the memory needed for training or inference.
  • Blender rendering: check your scene, textures and rendering engine against the GPU memory and software requirements.

How to rent a GPU server

  1. Choose your GPU model, review the pricing and sign in.
  2. Add account credit, then select your Linux image and server password.
  3. Follow the setup progress in your account. Rental usage billing starts when the server is ready.
  4. Connect using SSH, install your tools and back up your results before releasing the server.

Follow the SSH guide and application guides to connect and configure your workload.

Billing and access

Electricity is billed separately at CHF 0.32/kWh. Each operating system installation, including reinstalls, costs CHF 0.50.

Rental billing continues while the server is powered off and ends when the dashboard confirms release. Back up your files before release or reinstallation.

Standard Linux images provide SSH access. IPv6 is the default; shared IPv4 can be selected during deployment or enabled on your server page, when available, for CHF 0.50 per started 24 hours once verified working. Shared IPv4 provides an assigned TCP port range, not a dedicated IPv4 address; UDP and public ports 80/443 are not included.

Desktop applications and streaming require a separate setup. Windows arrangements and other fees are described on the pricing page.

SSH · Network specifications · PyTorch · Ollama · ComfyUI · Blender

All GPU models · Offers and availability API · Contact SwissGPU

Stay longer, save on compute

Choose an available prepaid term and pay less per hour where a discount is shown. One month is 30 days. Your compute price is fixed for the selected term.

GPUTermSavingCHF/hourCompute upfrontGet started
NVIDIA RTX 40901 months5%0.38CHF 273.60Choose server
NVIDIA RTX 40902 months10%0.36CHF 518.40Choose server
NVIDIA RTX 40903 months15%0.34CHF 734.40Choose server
NVIDIA RTX 40906 months20%0.32CHF 1382.40Choose server
NVIDIA RTX 409012 months25%0.30CHF 2592.00Choose server

Installation and metered electricity are separate, as detailed below. Keep wallet credit available for ongoing charges. The prepaid compute commitment is non-refundable on early release.

View as Markdown