96 GB of VRAM
Large image and video models, their text encoders and upscalers can stay loaded together instead of being swapped in and out of memory.
GPU servers · ComfyUI
Price list updated
ComfyUI is an open-source, node-based interface for building image, video and audio generation workflows. On a RyzenRDP GPU server it runs on a NVIDIA RTX PRO 6000 Blackwell SE with 96 GB of VRAM in Frankfurt, Germany: you install it over SSH, keep your models on the server's own disk, and open the interface in your browser through an SSH tunnel. Servers are billed monthly, from $1,951.09 a month.
Large image and video models, their text encoders and upscalers can stay loaded together instead of being swapped in and out of memory.
Queue a long batch or a video render and close your laptop. The server keeps working, and the results are there when you come back.
Checkpoints, LoRAs and video models add up quickly. The server's own disk holds them, and a 10 Gbps port helps when you download them.
ComfyUI has an HTTP API, so the workflow you built in the interface can be called from your own application.
People use ComfyUI with model families such as Stable Diffusion, FLUX and current video models. You download the models yourself and are responsible for following each model's license. This is your own installation on your own server, not a shared "ComfyUI online" service: nobody else's jobs are in your queue.
ComfyUI is not preinstalled. You install it from its repository, which takes a few commands. Sign in
with ssh root@YOUR_SERVER_IP and check the GPU first:
nvidia-smi
Then clone ComfyUI into a Python virtual environment:
apt update && apt install -y git python3-venv
git clone https://github.com/Comfy-Org/ComfyUI.git
cd ComfyUI
python3 -m venv venv && source venv/bin/activate
Install PyTorch for NVIDIA GPUs with the command given in the ComfyUI README, then the requirements, and start it:
pip install -r requirements.txt
python main.py
ComfyUI now listens on 127.0.0.1:8188 on the server. Put your checkpoints in models/checkpoints
and other model types in the matching folders under models/.
On your own computer run ssh -L 8188:127.0.0.1:8188 root@YOUR_SERVER_IP. It forwards your local port 8188 to ComfyUI on the server.
Go to http://127.0.0.1:8188. The ComfyUI interface loads as if it ran on your own machine, and the traffic travels inside the encrypted SSH connection.
Run ComfyUI inside tmux, or as a systemd service, so it keeps working after you disconnect.
ComfyUI has no sign-in of its own, and custom nodes run code on the server. If you start it with
--listen on a public address, anyone who finds the port can use your GPU and run workflows on your
machine. Keep it on 127.0.0.1 and use the SSH tunnel, or put it behind a VPN or a reverse proxy with
TLS and authentication. Install custom nodes only from sources you trust.
One GPU is the usual choice for ComfyUI: a workflow runs on a single GPU, and the single-GPU server has 16 vCPU, 128 GB RAM, 500 GB disk and a 10 Gbps port. Servers with more GPUs let you run several ComfyUI instances side by side, one per GPU.
| 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 follows free GPU capacity and is shown live: a row marked Out of stock cannot be ordered until capacity returns. If the plan you need is out of stock, ask the sales team and tell us which plan and how many servers you need. Everything about ordering and billing is on the GPU servers page.
Yes. ComfyUI runs on any Linux server with an NVIDIA GPU. On a RyzenRDP GPU server you install it over SSH from its repository and open the interface in your browser through an SSH tunnel.
No. The server runs Ubuntu 24.04 LTS with NVIDIA CUDA and gives you root access, and you install ComfyUI yourself. The install steps on this page take a few commands.
ComfyUI itself needs very little; the models you load decide. Small image models run in a few gigabytes, while large image and video models with their text encoders can use tens of gigabytes. A GPU with 96 GB leaves room to keep several of them loaded.
No. It is your own server with your own ComfyUI installation. You choose the version, the custom nodes and the models, and no one else shares your queue or your GPU memory.
They can share one interface through its queue, but ComfyUI has no user accounts. For separate workspaces, run one ComfyUI instance per person on different ports, or one per GPU on a multi-GPU server.
Yes. ComfyUI exposes an HTTP API on the same port as the interface. Export a workflow in API format and submit it from your application; keep that port reachable only from your application server.
No. GPU servers are billed monthly, from $1,951.09 a month. They suit steady production work rather than an occasional session.
NVIDIA RTX PRO 6000 Blackwell SE with 96 GB of VRAM in Frankfurt, Germany. From $1,951.09 a month.