Skip to content
/ ComfyUI-Intel-Arc-Clean-Install-Windows-venv-XPU- Public

Fully automated installation scripts for ComfyUI optimized for Intel Arc GPUs (A-Series) and Intel Core Ultra iGPUs with XPU backend, Triton acceleration, and GGUF quantized model support.

License

MIT license
115 stars 5 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

ai-joe-git/ComfyUI-Intel-Arc-Clean-Install-Windows-venv-XPU-

BranchesTags

Repository files navigation

ComfyUI Intel Arc GPU - Complete Installation Suite

Windows | Virtual Environment | XPU Backend | Triton Acceleration

Intel Arc PyTorch License

Fully automated installation scripts for ComfyUI optimized for Intel Arc GPUs (A-Series) and Intel Core Ultra iGPUs with XPU backend, Triton acceleration, and GGUF quantized model support.

🚀 Features

  • ✅ One-click installation - Automated setup with dependency resolution
  • ✅ Intel Arc optimized - Native XPU backend with PyTorch nightly builds
  • ✅ Triton acceleration - 6-11x faster GGUF model loading and inference
  • ✅ Isolated environment - Clean Python venv, no conflicts with other AI tools
  • ✅ Essential nodes included - ComfyUI-Manager, GGUF, VideoHelper, Impact Pack
  • ✅ Always up-to-date - Scripts pull latest ComfyUI and PyTorch versions
  • ✅ No manual patching - Automatic XPU detection and optimization

📋 Requirements

Hardware

Component Minimum Recommended
GPU Intel Arc A310 Intel Arc A770 16GB
iGPU Intel Core Ultra 5 Intel Core Ultra 7/9
RAM 16GB 32GB+
Storage 50GB free 100GB+ SSD

Software

  • Windows 10/11 (64-bit)
  • Python 3.10 or 3.11 - Download
  • Git for Windows - Download
  • Visual Studio Build Tools 2022 - Download
    • Required for Triton GGUF acceleration
    • Select: "Desktop development with C++"
  • Latest Intel Graphics Drivers - Download

🎯 Quick Start

1. Download Scripts

Clone this repository or download as ZIP:

git clone https://github.com/ai-joe-git/ComfyUI-Intel-Arc-Clean-Install-Windows-venv-XPU-.git
cd ComfyUI-Intel-Arc-Clean-Install-Windows-venv-XPU-

2. Run Installation (5 minutes)

Run scripts in this order:

1. INSTALL_ComfyUI_Intel_Arc_XPU.bat       # Core installation
2. INSTALL_Custom_Nodes.bat                 # Essential nodes
3. INSTALL_GGUF_Triton_Patch.bat           # GGUF acceleration
4. START_ComfyUI.bat                        # Launch ComfyUI

3. Access ComfyUI

Open your browser: http://127.0.0.1:8188

📦 What Gets Installed

Core Components

  • ComfyUI - Latest from official repository
  • PyTorch 2.11+ XPU Nightly - Intel Arc optimized builds
  • Triton XPU 3.6+ - GPU kernel acceleration
  • ComfyUI Frontend - Latest official UI

Essential Custom Nodes

  • ComfyUI-Manager - Node package manager
  • ComfyUI-GGUF - Quantized model support (Q4_0, Q8_0, etc.)
  • ComfyUI-VideoHelperSuite - Video generation tools
  • ComfyUI-Impact-Pack - Utility nodes
  • rgthree-comfy - Workflow optimization tools

Performance Optimizations

  • GGUF Triton Patch - Accelerated dequantization
    • Q4_0 models: ~11x faster
    • Q8_0 models: ~6x faster
    • Q4_1 models: ~8x faster

🔧 Detailed Installation Guide

Script 1: Core Installation

INSTALL_ComfyUI_Intel_Arc_XPU.bat

What it does:

  1. ✓ Verifies Python 3.10/3.11 and Git installation
  2. ✓ Checks for Visual Studio Build Tools (C++ compiler)
  3. ✓ Clones ComfyUI to C:\ComfyUI
  4. ✓ Creates isolated Python virtual environment
  5. ✓ Installs PyTorch XPU nightly builds
  6. ✓ Installs Triton XPU for acceleration
  7. ✓ Installs ComfyUI dependencies
  8. ✓ Verifies XPU device detection

Expected output:

PyTorch: 2.11.0.dev20260118+xpu
XPU available: True
Device: Intel(R) Arc(TM) A770 Graphics (16GB)

Script 2: Custom Nodes Installation

INSTALL_Custom_Nodes.bat

What it does:

  • Clones essential custom nodes to C:\ComfyUI\custom_nodes\
  • Installs node-specific dependencies
  • Updates existing nodes if already installed

Nodes installed:

  • ComfyUI-Manager (ltdrdata)
  • ComfyUI-GGUF (city96)
  • ComfyUI-VideoHelperSuite (Kosinkadink)
  • ComfyUI-Impact-Pack (ltdrdata)
  • rgthree-comfy (rgthree)

Script 3: GGUF Triton Patch

INSTALL_GGUF_Triton_Patch.bat

What it does:

  1. ✓ Verifies ComfyUI-GGUF node is installed
  2. ✓ Downloads latest Triton patch from this repo
  3. ✓ Applies patch to enable GPU-accelerated dequantization
  4. ✓ Verifies Triton kernels are active

Performance improvements:

Model Type Without Triton With Triton Speedup
Q4_0 GGUF Slow PyTorch Triton kernel ~11x faster
Q8_0 GGUF Slow PyTorch Triton kernel ~6x faster
Q4_1 GGUF Slow PyTorch Triton kernel ~8x faster
Q4_K_M PyTorch PyTorch No change*

*K-quants (Q4_K_M, Q5_K_M, Q6_K) not yet accelerated by this patch.

Script 4: Launch ComfyUI

START_ComfyUI.bat

What it does:

  • Initializes Visual Studio C++ environment for Triton
  • Sets Intel XPU environment variables
  • Activates Python virtual environment
  • Launches ComfyUI with optimized flags

Startup flags:

  • --lowvram - Efficient memory management for 8-16GB GPUs
  • --bf16-unet - BFloat16 precision (faster, lower VRAM)
  • --async-offload - Asynchronous model offloading
  • --disable-smart-memory - Predictable memory behavior

🎮 Supported Hardware

Intel Arc Discrete GPUs (Full Support ✅)

GPU Model VRAM Performance Notes
Arc A770 LE 16GB Excellent Best for video generation
Arc A770 8GB Very Good Recommended for most workflows
Arc A750 8GB Very Good Great price/performance
Arc A580 8GB Good Budget option
Arc A380 6GB Fair Entry level
Arc A310 4GB Limited Simple workflows only

Intel Core Ultra iGPUs (Supported ✅)

CPU Series iGPU Performance Notes
Core Ultra 9 Intel Arc iGPU Good Meteor Lake/Arrow Lake
Core Ultra 7 Intel Arc iGPU Good Best laptop option
Core Ultra 5 Intel Arc iGPU Fair Budget laptop

Intel Iris Xe (Experimental ⚠️)

  • 11th/12th Gen Intel Core with Iris Xe
  • Limited support, may fallback to CPU
  • Not recommended for production use

Legacy Intel Graphics (Not Supported ❌)

  • Intel UHD Graphics (10th Gen and older)
  • CPU-only mode (extremely slow)

🛠️ Updating ComfyUI

Run UPDATE_ComfyUI.bat to update:

  • ComfyUI core
  • PyTorch XPU nightly
  • Triton XPU
  • All custom nodes
  • Python dependencies

The script safely updates while preserving your models and workflows.

📊 Performance Benchmarks

LTX Video 2 (481 frames @ 768x512, 8 steps)

Hardware Time GGUF Triton Notes
Arc A770 16GB 25:32 Enabled Q8_0 FLUX + Qwen
Arc A770 8GB ~30:00 Enabled --lowvram required
Arc A750 8GB ~32:00 Enabled Comparable to A770 8GB
Core Ultra 7 ~45:00 Enabled iGPU only

FLUX.1 Dev (1024x1024, 20 steps)

Hardware Time VRAM Used
Arc A770 16GB ~45s 14GB
Arc A770 8GB ~60s 7.8GB (offloading)
Arc A750 8GB ~65s 7.8GB (offloading)

Benchmarks with GGUF Q8_0 models and Triton acceleration enabled.

🐛 Troubleshooting

Issue: "CUDA not available" or falls back to CPU

Solution:

# Verify XPU is detected
python -c "import torch; print(torch.xpu.is_available())"

If False:

  1. Update Intel Graphics drivers
  2. Reinstall PyTorch XPU: pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/xpu --force-reinstall

Issue: "Failed to find C++ compiler" error

Solution:

  1. Install Visual Studio Build Tools 2022
  2. Select "Desktop development with C++"
  3. Restart your PC
  4. Run START_ComfyUI.bat (not python main.py directly)

Issue: Out of Memory (OOM) errors

Solutions:

  • Use --lowvram flag (already in START script)
  • Try GGUF quantized models (Q4_0, Q8_0)
  • Reduce resolution or batch size
  • Close other GPU applications

Issue: Triton kernels not working

Verify Triton:

cd C:\ComfyUI
call comfyui_venv\Scripts\activate.bat
python -c "from custom_nodes.ComfyUI-GGUF.dequant import HAS_TRITON; print('Triton:', HAS_TRITON)"

If False:

pip install pytorch-triton-xpu --force-reinstall

Issue: Slow performance compared to expected

Checklist:

  • ✓ Triton patch applied? Check ComfyUI console for "Triton available, enabling optimized kernels"
  • ✓ Using GGUF Q8_0/Q4_0 models for acceleration?
  • ✓ GPU utilization at 100%? Check Task Manager
  • ✓ Power plan set to "High Performance"?
  • ✓ Latest Intel Graphics drivers installed?

💡 Tips for Best Performance

Model Format Recommendations

Use Case Model Format Why
Best Quality GGUF Q8_0 Minimal quality loss, 6x faster with Triton
Balanced GGUF Q4_K_M Good quality, smaller size
Maximum Speed GGUF Q4_0 11x faster with Triton, acceptable quality
Full Precision FP16/BF16 Highest quality, largest size, slowest

Memory Management

For 16GB Arc GPUs:

  • Remove --lowvram from START script for fastest performance
  • Can run most models without offloading

For 8GB Arc GPUs:

  • Keep --lowvram flag (default)
  • Use GGUF quantized models
  • Avoid loading multiple large models simultaneously

For 4-6GB Arc GPUs:

  • Add --novram for maximum offloading
  • Use Q4_0 GGUF models
  • Lower resolution workflows only

Workflow Optimization

  • Use GGUF models - Faster loading with Triton acceleration
  • Enable caching - Triton compiles kernels once, then caches
  • Batch processing - Process multiple frames/images together
  • Lower steps - 6-8 steps often sufficient with good samplers

📁 Directory Structure

After installation:

C:\ComfyUI\
├── comfyui_venv\           # Python virtual environment
├── custom_nodes\           # Custom nodes
│   ├── ComfyUI-Manager\
│   ├── ComfyUI-GGUF\       # Quantized models (Triton patched)
│   ├── ComfyUI-VideoHelperSuite\
│   ├── ComfyUI-Impact-Pack\
│   └── rgthree-comfy\
├── models\                 # Place models here
│   ├── checkpoints\
│   ├── clip\
│   ├── vae\
│   ├── loras\
│   └── unet\
├── input\                  # Input images/videos
├── output\                 # Generated outputs
├── user\                   # User settings
├── sycl_cache\            # XPU kernel cache
└── main.py                # ComfyUI entry point

🔗 Useful Resources

Official Documentation

Community

Model Resources

🤝 Contributing

Contributions welcome! Please:

  1. Test on Intel Arc hardware
  2. Document any issues or improvements
  3. Submit PR with clear description

📜 License

MIT License - See LICENSE for details

🙏 Credits

  • Scripts: ai-joe-git
  • ComfyUI: comfyanonymous
  • GGUF Node: city96
  • Intel XPU Community: Everyone testing and sharing knowledge

⭐ Support

If these scripts helped you, please:

  • ⭐ Star this repository
  • 🐛 Report issues on GitHub
  • 💬 Share your results in Discussions
  • 📢 Help other Intel Arc users!

Last Updated: January 2026
ComfyUI Version: 0.9.2+
PyTorch XPU Version: 2.11.0.dev+
Tested Hardware: Arc A770, A750, A580, Core Ultra 7/9

🚀 What's New

January 2026

  • ✅ Triton XPU integration for GGUF acceleration
  • ✅ Automated patch installer with GitHub download
  • ✅ Visual Studio Build Tools detection
  • ✅ PyTorch 2.11+ nightly XPU builds
  • ✅ Streamlined installation process
  • ✅ Performance improvements for Q8_0/Q4_0 models

Coming Soon

  • 🔄 K-quant Triton kernels (Q4_K_M, Q5_K_M)
  • 🔄 Automatic model downloader
  • 🔄 ComfyUI Portable build option
  • 🔄 Docker container for Intel Arc

🔧 Repair/Update PyTorch XPU

If you experience issues with PyTorch or want to update to the latest nightly build:

When to use:

  • PyTorch XPU not detecting your Intel Arc GPU
  • After updating Intel Graphics drivers
  • ComfyUI showing "Device: cpu" instead of "xpu"
  • Upgrading to latest PyTorch nightly build
  • Fixing corrupted PyTorch installation

How to use:

  1. Close ComfyUI if running
  2. Run REPAIR_PyTorch_XPU.bat
  3. Wait for installation to complete (~5-10 minutes)
  4. Verify XPU is detected in the output
  5. Run START_ComfyUI.bat to test

Expected Output:

PyTorch Version: 2.11.0.dev20260119+xpu XPU Available: True GPU Device: Intel(R) Arc(TM) A770 Graphics GPU Count: 1 Triton Version: 3.6.0

text

If you see this, PyTorch XPU is working correctly! ✅

Made with ❤️ for the Intel Arc community

About

Fully automated installation scripts for ComfyUI optimized for Intel Arc GPUs (A-Series) and Intel Core Ultra iGPUs with XPU backend, Triton acceleration, and GGUF quantized model support.

Resources

Readme

License

MIT license
Activity

Stars

115 stars

Watchers

7 watching

Forks

5 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages