In this guide, we will introduce the system requirements for installing ComfyUI. Due to frequent updates of ComfyUI, this document may not be updated in a timely manner. Please refer to the relevant instructions in ComfyUI.

Regardless of which version of ComfyUI you use, it runs in a separate Python environment.

You can refer to the following sections to learn about the installation methods for different systems and versions of ComfyUI. In the installation of different versions, we have simply described the system requirements.

​
Nvidia 50 Series GPU Requirements

To make your Nvidia 50 series GPU (Blackwell architecture) work properly with ComfyUI, you need a PyTorch version that supports CUDA 12.8 or newer. Currently (March 2025), the stable version of PyTorch does not yet support the Blackwell architecture, so you need to use a nightly build version.

Related discussions on this issue are concentrated here.

​
For Windows Users

Recommended Option: Download the standalone ComfyUI Portable version with nightly pytorch 2.7 cu128:

Other Options: Older torch 2.6 Windows package:

​
Manual Installation

Windows and Linux users can install the PyTorch nightly version using the following command:

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

​
Docker Container Alternative

You can try the PyTorch container provided by Nvidia, which might offer better performance.

Container address: https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch

Usage method:

docker run -p 8188:8188 --gpus all -it --rm nvcr.io/nvidia/pytorch:25.01-py3

Inside the Docker container, execute the following commands:

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
grep -v 'torchaudio\|torchvision' requirements.txt > temp_requirements.txt
pip install -r temp_requirements.txt
python main.py --listen