A PowerShell script that sets up ComfyUI with ROCm support for AMD RDNA GPUs on Windows. It uses uv for fast and reliable Python package management.
- Native AMD GPU support (RDNA 3, 3.5, and 4)
- Fast installs using uv instead of pip
- One-command setup
- GPU configuration is saved between runs
- Handles venvs, dependencies, and ROCm PyTorch automatically
| Generation | GPUs | Architecture |
|---|---|---|
| RDNA 3 | RX 7900 XTX, RX 7900 XT, RX 7800 XT, RX 7700 XT, RX 7600 | gfx1100 |
| RDNA 3.5 | Ryzen AI (Strix Point / Halo), Radeon 890M, 8060S | gfx1151 |
| RDNA 4 | RX 9070 XT, RX 9070 | gfx1200 |
- Windows 10 or 11 (64-bit)
- Git (download)
- PowerShell 5.1 or newer (already included with Windows)
Open PowerShell and run:
irm https://raw.githubusercontent.com/SystemVll/ComfyUI-AMD/main/ComfyUI-AMD.ps1 | iex-
Clone the repository:
git clone https://github.com/SystemVll/ComfyUI-AMD.git cd ComfyUI-AMD
-
Run the script:
.\ComfyUI-AMD.ps1
- Installs uv if it isn’t already installed
- Clones the ComfyUI repository
- Creates a Python 3.12 virtual environment
- Installs the correct ROCm PyTorch build for your GPU
- Installs ComfyUI dependencies
- Configures HIP_VISIBLE_DEVICES
- Launches the ComfyUI web UI
On the first run, you’ll be asked which GPU generation you’re using:
1– RDNA 3 (RX 7000 series)2– RDNA 3.5 (Ryzen AI)3– RDNA 4 (RX 9000 series)
Your choice is saved and reused on future runs.
If more than one GPU is detected, you’ll be prompted to set HIP_VISIBLE_DEVICES:
0– first GPU1– second GPU0,1– use both
your-folder/
├── setup.ps1
├── .gpu_config.json
└── ComfyUI/
├── .venv/
├── models/
├── custom_nodes/
└── output/
.\setup.ps1Follow the prompts and ComfyUI will start automatically.
.\setup.ps1You can usually just press Enter — your settings are reused.
Once running, open:
http://127.0.0.1:8188
cd ComfyUI
git pull
cd ..
.\setup.ps1Any new dependencies will be handled automatically.
uv is a modern Python package manager written in Rust. Compared to pip, it’s significantly faster and tends to be more reliable, especially for large dependency trees.
- Restart PowerShell and try again
Set-ExecutionPolicy RemoteSigned -Scope CurrentUser- Update your AMD drivers
- Verify your GPU supports ROCm
- Try setting
HIP_VISIBLE_DEVICES=0manually
- Make sure port 8188 isn’t already in use
- Check the console output for errors
- Delete the
.venvfolder and rerun the script
- ComfyUI: https://github.com/comfyanonymous/ComfyUI
- ROCm: https://rocm.docs.amd.com/
- uv: https://docs.astral.sh/uv/
- Models: https://huggingface.co/models
- First install can take a while, especially on slower connections
- ROCm PyTorch wheels are large (2–3 GB)
- Models are not included — place them in
ComfyUI/models/
Issues and pull requests are welcome.
This script is provided as-is. ComfyUI is licensed separately; see the ComfyUI repository for details.