Quick start guide for running ACE-Step on Intel Arc GPUs and integrated graphics.
- Intel Arc GPU: A770, A750, A580, A380, or Intel integrated graphics
- Python 3.11: Download from python.org
- Latest Intel GPU drivers: Install from Intel's website
- Internet connection: For first-time setup
- Disk space: ~5-10 GB for dependencies
-
Run the setup script:
setup_xpu.bat
-
Wait for installation (takes a few minutes on first run)
-
Launch the Gradio UI:
start_gradio_ui_xpu.bat
-
Open your browser: http://127.0.0.1:7860
-
Create virtual environment:
python -m venv venv_xpu
-
Activate it:
call venv_xpu\Scripts\activate -
Upgrade pip:
python -m pip install --upgrade pip
-
Install XPU dependencies:
pip install -r requirements-xpu.txt
-
Launch:
start_gradio_ui_xpu.bat
Models should be placed in the checkpoints folder. If you already have models from a previous installation, they will be automatically detected.
- DiT Model:
acestep-v15-turbo(fast generation) - LM Model:
acestep-5Hz-lm-4B(best quality, uses CPU offload)
-
Automatic (uses defaults):
start_gradio_ui_xpu.bat
-
Manual (choose models interactively):
start_gradio_ui_xpu_manual.bat
-
API Server (REST API access):
start_api_server_xpu.bat
The bat files automatically set these performance optimizations:
set SYCL_CACHE_PERSISTENT=1
set SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1
set PYTORCH_DEVICE=xpu
set TORCH_COMPILE_BACKEND=eagerThese settings improve XPU performance and are based on verified working configurations.
Create a .env file in the root directory to customize settings:
# Gradio UI Settings
PORT=7860
SERVER_NAME=127.0.0.1
LANGUAGE=en
# Model Settings
ACESTEP_CONFIG_PATH=acestep-v15-turbo
ACESTEP_LM_MODEL_PATH=acestep-5Hz-lm-4B
ACESTEP_OFFLOAD_TO_CPU=true
# API Settings
ACESTEP_API_KEY=your-secret-keyRun setup_xpu.bat to create the virtual environment.
- Check that your GPU drivers are up to date
- Verify PyTorch XPU installation:
call venv_xpu\Scripts\activate python -c "import torch; print(torch.xpu.is_available())"
Reinstall PyTorch XPU:
call venv_xpu\Scripts\activate
pip uninstall torch torchaudio torchvision
pip install --pre torch torchaudio torchvision --index-url https://download.pytorch.org/whl/nightly/xpu- Use a smaller LM model (0.6B or 1.7B instead of 4B)
- Enable CPU offload in the UI
- Close other GPU-intensive applications
- MP3/Opus/AAC files use torchaudio with ffmpeg backend (bundled)
- FLAC/WAV files use soundfile (fastest)
- If issues occur, try converting to WAV format
| Script | Description |
|---|---|
setup_xpu.bat |
One-command environment setup |
start_gradio_ui_xpu.bat |
Launch Gradio web UI (automatic) |
start_gradio_ui_xpu_manual.bat |
Launch Gradio UI with model selection |
start_api_server_xpu.bat |
Launch REST API server |
- ✅ WAV/FLAC: Native support via soundfile (fastest)
- ✅ MP3: Supported via torchaudio with ffmpeg backend
- ✅ Opus/AAC: Supported via torchaudio
No additional codec installation needed!
- First launch takes longer as models are initialized
- CPU offload is recommended for 4B LM on GPUs with <=16GB VRAM
- torch.compile is disabled (not fully supported on XPU yet)
- Python 3.11 is recommended for best compatibility
- Check the main documentation:
README.md - Verify XPU installation: Run verification commands above
- Update check: Bat files automatically check for updates on startup
Once setup is complete, simply run:
start_gradio_ui_xpu.batAnd start creating music with ACE-Step on your Intel GPU!