Whole GPUs, 96 GB each
A plan lists 1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs. Each GPU comes with its full 96 GB of VRAM.
GPU servers · Frankfurt, Germany
Price list updated
A RyzenRDP GPU server is a Linux server in Frankfurt, Germany with 1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs and 96 GB of VRAM per GPU. The single-GPU plan has 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port and costs $1,951.09 a month; the largest plan has 8 GPUs with 768 GB of VRAM in total. Every server runs Ubuntu 24.04 LTS with NVIDIA CUDA, gives you root access over SSH, is billed monthly, and can be ordered when GPU capacity is free.
Four plans, each with NVIDIA RTX PRO 6000 Blackwell SE GPUs. A plan's vCPU, RAM and disk grow with its GPU count, and every plan has a 10 Gbps port and 20 TB of outbound transfer a month. Prices are per month in US dollars.
| GPUs | VRAM per GPU | Total VRAM | vCPU | RAM | Disk | Monthly | Stock and order |
|---|---|---|---|---|---|---|---|
| 1 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 96 GB | 16 | 128 GB | 500 GB | $1,951.09 | |
| 2 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 192 GB | 32 | 256 GB | 1,000 GB | $3,900.60 | |
| 4 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 384 GB | 64 | 512 GB | 2,000 GB | $7,799.63 | |
| 8 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | 768 GB | 128 | 1,024 GB | 4,000 GB | $15,597.68 |
Stock is live. GPU servers depend on free GPU capacity, which is not always there. A row marked Out of stock cannot be ordered until capacity returns; a row with an Order button can be ordered now. If the plan you need is out of stock, ask the sales team and tell us which plan and how many servers you need: we will tell you what we know about when they can be built.
A plan lists 1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs. Each GPU comes with its full 96 GB of VRAM.
Servers run Ubuntu 24.04 LTS with NVIDIA CUDA. You sign in as root over SSH and install the frameworks and models you want.
A 10 Gbps port and 20 TB of outbound transfer a month, which matters when you pull model files of tens of gigabytes.
From 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port on the single-GPU plan to 128 vCPU, 1,024 GB RAM, 4,000 GB disk and a 10 Gbps port on the largest.
Every GPU server runs in Frankfurt. Your models, prompts and outputs are stored and processed on that server.
One price per month, shown in the table. Pay by cryptocurrency or a credit or debit card. There is no hourly metering to watch.
Serve an open-weight language model behind an OpenAI-compatible API with Ollama or vLLM.
Keep prompts and documents on a server you control instead of sending them to a model provider's API.
Run image and video generation workflows in ComfyUI and reach the interface from your browser.
Ollama, vLLM, Open WebUI, ComfyUI, PyTorch and JupyterLab: what each does and how to install it.
The same servers suit fine-tuning and LoRA training, embedding and reranking models for retrieval, speech-to-text, and GPU rendering. What they have in common is that the work needs a lot of VRAM on hardware you control. The GPU itself is described on the NVIDIA RTX PRO 6000 Blackwell page.
Choose a GPU plan by total VRAM: the model weights, the context cache and the framework's own overhead all have to fit in it. A model's weights take roughly its parameter count times the bytes stored per parameter, so a 70-billion-parameter model needs about 35 GB at 4-bit and about 140 GB at 16-bit.
| Plan | Total VRAM | Monthly price | What fits as an estimate |
|---|---|---|---|
| 1 × NVIDIA RTX PRO 6000 Blackwell SE | 96 GB | $1,951.09 | One model with up to about 70 billion parameters at 4-bit or 8-bit, with room left for context; image and video generation; LoRA fine-tuning |
| 2 × NVIDIA RTX PRO 6000 Blackwell SE | 192 GB | $3,900.60 | A 70-billion-parameter model at 16-bit split across both GPUs, or two different models side by side |
| 4 × NVIDIA RTX PRO 6000 Blackwell SE | 384 GB | $7,799.63 | Models beyond 100 billion parameters at higher precision, several models at once, or more users on one model |
| 8 × NVIDIA RTX PRO 6000 Blackwell SE | 768 GB | $15,597.68 | Multi-GPU fine-tuning, and models of several hundred billion parameters at 8-bit or 16-bit |
These fits are estimates from the size of the weights alone. Long contexts and many simultaneous users add to the cache, so leave headroom. The LLM hosting page works through the arithmetic by model size and precision.
These plans are virtual servers with whole GPUs. Each plan lists its vCPU, RAM and disk like any virtual server, and adds 1, 2, 4 or 8 GPUs that come with their full 96 GB of VRAM. That is what most people mean by a GPU VPS, a cloud GPU server or a GPU instance: you get root on your own Linux system, and the machine is created for you instead of being racked for you.
It is not a bare-metal machine. If you need a physical dedicated GPU server built to your own specification, in a data center you choose, that is an enterprise quote rather than one of these plans.
And it is billed by the month, not by the hour. A GPU server here is for work that runs for weeks: an inference endpoint your product depends on, a private assistant for a team, a render or training queue.
Choose a plan in the table above by total VRAM. A row with an Order button can be ordered now.
The order form opens in the RyzenRDP console. Pay the first month by cryptocurrency or a credit or debit card.
When the server is built, its IP address and root password appear in the console. Connect with ssh root@YOUR_SERVER_IP.
Run nvidia-smi. It lists every GPU of the plan with its memory, and it is the first thing to check before you install anything.
Install the runtime you need: Ollama or vLLM for language models, ComfyUI for image and video workflows. The apps page has the commands.
Often not. An AI agent, a chatbot or an n8n workflow that calls a hosted model through an API does its heavy work on the model provider's hardware. The server only builds the prompt and handles the reply, which is ordinary CPU work. For that, an AI VPS on CPU from $14.99 a month is the right size.
You need a GPU server when the model itself runs on your server: a self-hosted LLM, image or video generation, training or fine-tuning.
A GPU server is a server with one or more graphics processors that run AI models, rendering and other highly parallel work much faster than a CPU can. A RyzenRDP GPU server is a Linux server in Frankfurt, Germany with 1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs, each with 96 GB of VRAM, and root access over SSH.
The NVIDIA RTX PRO 6000 Blackwell SE, NVIDIA's data center card of the Blackwell generation with 96 GB of memory. Plans have 1, 2, 4 or 8 of them. See the RTX PRO 6000 Blackwell page for what NVIDIA publishes about the card.
From $1,951.09 to $15,597.68 a month. The single-GPU plan, with 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port, costs $1,951.09 a month, and the two-GPU plan costs $3,900.60 a month. Every price is in the plan table.
No. GPU servers are billed monthly. Spread over the 730 hours of an average month, the single-GPU plan works out to about $2.67 per GPU-hour, but you pay for the whole month. They suit workloads that run all month, such as an inference endpoint or a private assistant, rather than a job of a few hours.
Stock follows free GPU capacity and changes. The plan table shows it live: a row marked Out of stock cannot be ordered until capacity returns. Ask the sales team if the plan you need is out of stock.
GPU servers run Ubuntu 24.04 LTS with NVIDIA CUDA. Sign in as root over SSH and run nvidia-smi to see the GPUs. Windows is not available on GPU servers.
It is a virtual server with whole GPUs: the plan lists its vCPU, RAM and disk, and 1, 2, 4 or 8 GPUs that each come with their full 96 GB of VRAM. It is not a bare-metal machine. For physical dedicated GPU hardware, ask for an enterprise quote.
Roughly the size of the model's weights plus headroom for context. Weights take about the parameter count times the bytes per parameter: a 70-billion-parameter model is about 35 GB at 4-bit and about 140 GB at 16-bit. One GPU has 96 GB; the largest plan has 768 GB in total. The LLM hosting page has a sizing table.
In Frankfurt, Germany. Every GPU plan runs there.
No. Cryptocurrency mining of any kind is prohibited by our acceptable use policy.
By cryptocurrency or a credit or debit card, one month at a time. Paying with crypto works the same way as for every other plan.
1, 2, 4 or 8 NVIDIA RTX PRO 6000 Blackwell SE GPUs, 96 GB of VRAM each, Ubuntu 24.04 LTS with NVIDIA CUDA. From $1,951.09 a month.