discussion Native ROCm on Windows: Running ComfyUI on AMD without ZLuda

[Fallout4ever]

[GUIDE] Native ROCm for ComfyUI on Windows (No ZLuda)​

---

What and Why​

Most of the current documentation for AMD cards on Windows still points toward ZLuda or dual-booting Linux. When guides aren't written for those specific use cases, they’re usually written for power users who are already comfortable with troubleshooting.
Since ZLuda development has stalled and native ROCm support is finally functional on Windows, I’ve spent the last year refining a stable setup. My goal is to break this down so that even if you’ve never touched a command line, you can get this running. This is a direct method to get ComfyUI working natively on RDNA hardware without translation layers, virtual machines, or WSL2.

---

Terminology​

If you’re new to this, the acronyms can be a wall. Here is the plain-English breakdown of what we’re actually dealing with:

• ROCm: This is AMD’s "AI engine." Think of it as the software that tells your graphics card how to do the heavy math required for AI. For years, it was Linux-only, but it's finally stable on Windows.
• CUDA: This is the NVIDIA version of ROCm. Most AI tools were built for CUDA first, which is why AMD users have had to use workarounds for so long.
• ZLuda: A "translator" layer. It tries to trick Windows into thinking your AMD card is actually an NVIDIA card so it can run CUDA code. It worked for a while, but it's essentially a dead end now that native support exists.
• WSL2 (Windows Subsystem for Linux): Basically a "Linux window" inside Windows. It’s powerful, but it’s a massive headache for beginners because it handles files differently and eats up a lot of system resources. We are skipping this entirely.
• Git: A tool used to download and update code from the internet (specifically from a site called GitHub). Think of it as a specialized downloader that ensures you always have the latest version of the software.
• ComfyUI: The actual program we’re trying to run. It’s a node-based interface that gives you a lot more control over image generation than basic "type and click" apps.
• Python: The programming language ComfyUI is written in. You don't need to know how to code, but you do need to have the right version installed for the "engine" to work.
• Venv (Virtual Environment): Think of this as an isolated "sandbox" for a specific project. AI software is notorious for needing very specific versions of different tools. A venv ensures that the stuff we install for ComfyUI stays inside the ComfyUI folder and doesn't mess with anything else on your computer.
  
  • Command Prompt is the "old" way—it's basic and has been around since the 80s.
  • PowerShell is the modern version. It’s more powerful and handles the complex scripts we need for ROCm and Python much more reliably.
• PIP: This is the "package manager" for Python. It’s essentially a command-line App Store. We use it to tell the computer, "Go find this specific piece of AI software and install it for me."

---

System Requirements​

Before you start downloading, you need to make sure your hardware can actually handle the workload. AI generation is extremely taxing on the GPU (Video Card) and VRAM (Video Memory).

1. Supported AMD Generations​

ROCm on Windows currently focuses on the most recent architectures. To run this natively, you generally need an RDNA 2 or RDNA 3 (and now RDNA 4) card.

• RDNA 3 & 4 (The "Easy" Tier): Cards like the RX 7900 XTX, 7800 XT, or the newer 9000 series. These have the most robust native support.
• RDNA 2 (The "Workhorse" Tier): Cards like the RX 6950 XT, 6800, or 6800 XT. While officially supported, these sometimes require an "override" (a small settings tweak) to tell the software to treat them like the newer cards.
• Older Cards (RDNA 1 / GCN): If you have an RX 5000 series or older (like an RX 580), this native ROCm method likely won't work for you yet.

2. VRAM: The "Gas Tank" for AI​

VRAM is the memory built into your graphics card.

• 8GB: The bare minimum. You can run basic models (SD1.5), but you’ll struggle with newer, high-detail ones like SDXL or Flux.
• 12GB - 16GB: The "Sweet Spot." This allows you to run almost everything comfortably at decent speeds.
• 24GB+: Professional grade. Necessary if you want to train your own models or run massive "Cloud-Class" AI locally.

3. How to check what you have​

If you aren't sure exactly which model or how much VRAM you have, Windows makes it easy to check:

1. Press Ctrl + Shift + Esc to open the Task Manager.
2. Click on the Performance tab on the left.
3. Scroll down to GPU at the bottom of the list.
4. In the top right, it will show your model (e.g., "AMD Radeon RX 7900 XTX").
5. Look at Dedicated GPU Memory at the bottom—that is your total VRAM (e.g., "24.0 GB").

---

Requirement	Minimum	Recommended
GPU Generation	RDNA 2 (RX 6000)	RDNA 3 or 4 (RX 7000/9000)
VRAM	8 GB	16 GB+
System RAM	16 GB	32 GB+
OS	Windows 10/11	Windows 11 (Latest Build)

---

First Steps: The Foundation
To get ComfyUI running, we need to install a few core tools. We’ll be using PowerShell for most of this. If you’ve never used it, don’t worry—it’s just a window where you type (or paste) commands to tell your computer exactly what to do.

1. Install the latest version of Adrenaline drivers​

2. Install Git. Git is the tool that "clones" (downloads) the ComfyUI code and keeps it updated.

1. Option A (Fastest): Paste this into PowerShell and hit Enter: winget install --id Git.Git -e --source winget
2. Option B (Manual): Download the "64-bit Git for Windows Setup" from the
   You must be registered to see links
   .

3. Install Python 3.12 Python is the engine that runs ComfyUI. We are using version 3.12 because it is currently the most stable "sweet spot" for AI dependencies on Windows.

• Option A (Fastest): Paste this into PowerShell and hit Enter: winget install --id Python.Python.3.12 -e
• Option B (Manual): Download the "Windows installer (64-bit)" from the
  You must be registered to see links
  .
  
  CRITICAL: If you install manually, make sure you check the box that says "Add Python to PATH" at the start of the installer. If you miss this, nothing else will work.

Note: After installing Python and Git, close your PowerShell window and open a new one. This "refreshes" the system so it recognizes the new tools you just added.

4. Download ComfyUI Now we actually grab the software. We’re going to create a folder for it and download the code.

1. Decide where you want ComfyUI to live (e.g., your C: drive or a secondary SSD). In PowerShell, navigate there. For example, to go to your C: drive: cd C:\ or if you want it to be in a folder called AI: cd C:\AI\
2. Run the clone command: git clone https://github.com/comfy-org/ComfyUI.git
3. Move into the new folder: cd ComfyUI or cd AI\ComfyUI

---

Setting Up the Environment​

Now that we have the files, we need to create the "sandbox" (Venv) and install the actual AI engine.

1. Create the Virtual Environment While in your ComfyUI folder in PowerShell, run this: python -m venv venv
2. Activate the Environment You must "enter" the sandbox before installing anything. You’ll know it worked because (venv) will appear at the start of your PowerShell line. \venv\Scripts\Activate.ps1

---

Install PyTorch via PIP​

Full disclaimer, this part is lifted directly from AMD’s own website. Warning, this will take several minutes.

You must be registered to see links

Enter the commands to set up ROCm environment:

pip install --no-cache-dir `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/rocm_sdk_core-7.2.0.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/rocm_sdk_devel-7.2.0.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/rocm_sdk_libraries_custom-7.2.0.dev0-py3-none-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/rocm-7.2.0.dev0.tar.gz

Enter the commands to install torch, torchvision and torchaudio for ROCm AMD GPU support:

pip install --no-cache-dir `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/torch-2.9.1%2Brocmsdk20260116-cp312-cp312-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/torchaudio-2.9.1%2Brocmsdk20260116-cp312-cp312-win_amd64.whl `
    https://repo.radeon.com/rocm/windows/rocm-rel-7.2/torchvision-0.24.1%2Brocmsdk20260116-cp312-cp312-win_amd64.whl

---

Verify PyTorch installation​

Confirm if PyTorch is correctly installed.
Verify if Pytorch is installed and detecting the GPU compute device: python -c "import torch" 2>nul && echo Success || echo Failure
Expected result: Success
Enter command to test if the GPU is available: python -c "import torch; print(torch.cuda.is_available())"
Expected result: True
Enter command to display installed GPU device name: python -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))"
Example result: device name [0]: Radeon RX 7900 XTX
Enter command to display component information within the current PyTorch environment: python -m torch.utils.collect_env
Example result:

PyTorch version: 2.9.1+rocmsdk20260116​

Is debug build: False​

CUDA used to build PyTorch: N/A​

ROCM used to build PyTorch: 7.2.26024-f6f897bd3d​

​

OS: Microsoft Windows 11 Pro (10.0.26100 64-bit)​

GCC version: Could not collect​

Clang version: Could not collect​

CMake version: Could not collect​

Libc version: N/A​

​

Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)] (64-bit runtime)​

Python platform: Windows-11-10.0.26100-SP0​

Is CUDA available: True​

CUDA runtime version: Could not collect​

CUDA_MODULE_LOADING set to:​

GPU models and configuration: AMD Radeon PRO W7900 (gfx1100)​

Nvidia driver version: Could not collect​

cuDNN version: Could not collect​

Is XPU available: False​

HIP runtime version: 7.2.26024​

MIOpen runtime version: 3.5.1​

Is XNNPACK available: True​

---

Install ComfyUI Dependencies:​

pip install -r requirements.txt

---

How to Launch​

Whenever you want to run ComfyUI:

1. Open PowerShell in your ComfyUI folder.
2. Run .\venv\Scripts\Activate.ps1
3. Run python main.py

---

This is my first time posting on here, let alone a forum in general, so I'm sure my formating is terrible. Feel free to give feedback, and any advice is much appreciated.