Skip to content

README.adoc

Latest commit

 

History

History
192 lines (149 loc) · 6.22 KB

README.adoc

File metadata and controls

192 lines (149 loc) · 6.22 KB

Run ComfyUI with ROCm on AMD GPU

Note: Image Building

This Docker image is often too big to build on GitHub Actions (throw "No space left on device" error). So the commands below contain the steps for building Docker image (basically downloading packages).

You can skip those steps if the rocm6 image on Docker Hub is recently built.

Prepare

Build & Run

You may need to add these configuration (especially for APUs) into the command of docker run, podman run below. (Credit to nhtua)

  • For RDNA 2 cards:

    • -e HSA_OVERRIDE_GFX_VERSION=10.3.0 \

  • For RDNA 3 cards:

    • -e HSA_OVERRIDE_GFX_VERSION=11.0.0 \

    • Check the AMD doc to see if your GPU can use 11.0.1.

  • For RDNA 4 cards:

    • -e HSA_OVERRIDE_GFX_VERSION=12.0.0 \

    • Check the AMD doc to see if your GPU can use 12.0.1.

  • For integrated graphics on CPU:

    • -e HIP_VISIBLE_DEVICES=0 \

You may also want to add more environment variable(s):

  • Enable tunable operations (slow first run, but faster subsequent runs. Doc1, Doc2). (Thanks to SergeyFilippov)

    • -e PYTORCH_TUNABLEOP_ENABLED=1 \

With Docker
# Build the image
git clone https://github.com/YanWenKun/ComfyUI-Docker.git
cd ComfyUI-Docker/rocm6
docker build . -t yanwk/comfyui-boot:rocm6

# Run the container
mkdir -p \
  storage \
  storage-models/models \
  storage-models/hf-hub \
  storage-models/torch-hub \
  storage-user/input \
  storage-user/output \
  storage-user/workflows

docker run -it --rm \
  --name comfyui-rocm6 \
  --device=/dev/kfd --device=/dev/dri \
  --group-add=video --ipc=host --cap-add=SYS_PTRACE \
  --security-opt seccomp=unconfined \
  --security-opt label=disable \
  -p 8188:8188 \
  -v "$(pwd)"/storage:/root \
  -v "$(pwd)"/storage-models/models:/root/ComfyUI/models \
  -v "$(pwd)"/storage-models/hf-hub:/root/.cache/huggingface/hub \
  -v "$(pwd)"/storage-models/torch-hub:/root/.cache/torch/hub \
  -v "$(pwd)"/storage-user/input:/root/ComfyUI/input \
  -v "$(pwd)"/storage-user/output:/root/ComfyUI/output \
  -v "$(pwd)"/storage-user/workflows:/root/ComfyUI/user/default/workflows \
  -e CLI_ARGS="" \
  yanwk/comfyui-boot:rocm6
With Podman
# Build the image
git clone https://github.com/YanWenKun/ComfyUI-Docker.git
cd ComfyUI-Docker/rocm6
podman build . -t yanwk/comfyui-boot:rocm6

# Run the container
mkdir -p \
  storage \
  storage-models/models \
  storage-models/hf-hub \
  storage-models/torch-hub \
  storage-user/input \
  storage-user/output \
  storage-user/workflows

podman run -it --rm \
  --name comfyui-rocm6 \
  --device=/dev/kfd --device=/dev/dri \
  --group-add=video --ipc=host --cap-add=SYS_PTRACE \
  --security-opt seccomp=unconfined \
  --security-opt label=disable \
  -p 8188:8188 \
  -v "$(pwd)"/storage:/root \
  -v "$(pwd)"/storage-models/models:/root/ComfyUI/models \
  -v "$(pwd)"/storage-models/hf-hub:/root/.cache/huggingface/hub \
  -v "$(pwd)"/storage-models/torch-hub:/root/.cache/torch/hub \
  -v "$(pwd)"/storage-user/input:/root/ComfyUI/input \
  -v "$(pwd)"/storage-user/output:/root/ComfyUI/output \
  -v "$(pwd)"/storage-user/workflows:/root/ComfyUI/user/default/workflows \
  -e CLI_ARGS="" \
  yanwk/comfyui-boot:rocm6

Once the app is loaded, visit http://localhost:8188/

ROCm: If you want to dive in…​

(Just side notes. Nothing to do with this Docker image)

The commands below use the AMD prebuilt ROCm PyTorch image.

This image is large in filesize. But if you have hard time to run the container, it may be helpful. As it takes care of PyTorch, the most important part, and you just need to install few more Python packages in order to run ComfyUI.

docker pull rocm/pytorch:rocm6.4.4_ubuntu24.04_py3.12_pytorch_release_2.7.1

mkdir -p storage

docker run -it --rm \
  --name comfyui-rocm6 \
  --device=/dev/kfd --device=/dev/dri \
  --group-add=video --ipc=host --cap-add=SYS_PTRACE \
  --security-opt seccomp=unconfined \
  --security-opt label=disable \
  -p 8188:8188 \
  --user root \
  --workdir /root/workdir \
  -v "$(pwd)"/storage:/root/workdir \
  rocm/pytorch:rocm6.4.4_ubuntu24.04_py3.12_pytorch_release_2.7.1 \
  /bin/bash

git clone https://github.com/comfyanonymous/ComfyUI.git

pip install -r ComfyUI/requirements.txt
# Or:
# conda install --yes --file ComfyUI/requirements.txt

python ComfyUI/main.py --listen --port 8188
# Or:
# python3 ComfyUI/main.py --listen --port 8188

Additional notes for Windows users

(Just side notes. Nothing to do with this Docker image)

WSL2 supports ROCm and DirectML:

  • ROCm

  • DirectML

  • ZLUDA

    • This is not using WSL2, it’s running natively on Windows. ZLUDA can "translate" CUDA codes to run on AMD GPUs. But as the first step, I recommend to try running SD-WebUI with ZLUDA, it’s easier to start with.