Docker image with AMD support for AUTOMATIC1111/stable-diffusion-webui
Install amdgpu driver with ROCm 6.0.2 support. Download at https://www.amd.com/en/support/linux-drivers
Instructions adapted from https://amdgpu-install.readthedocs.io/en/latest/install-installing.html
sudo apt install ./amdgpu-install_6.0.60002-1_all.deb
amdgpu-install -y --accept-eula --usecase=rocm
sudo usermod -a -G render $LOGNAME
sudo usermod -a -G video $LOGNAME
Reboot for changes to take effect.
pytorch is currently built for gfx90a, gfx900, gfx906, gfx908, gfx1030, gfx1100. You can check what processor your GPU is by either visiting LLVM's documentation (https://llvm.org/docs/AMDGPUUsage.html#id16) or by running the following command:
rocminfo | grep gfx
If your processor is not built by pytorch, you will need to provide the HSA_OVERRIDE_GFX_VERSION environment variable with the closet version. For example, an RX 67XX XT has processor gfx1031 so it should be using gfx1030. To use gfx1030, set HSA_OVERRIDE_GFX_VERSION=10.3.0 in docker-compose.yml.
For some cards, you will need to provide extra arguments in docker-compose.yml (some of which will boost performance) https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs
For many AMD GPUs, you must add --precision full --no-half or --upcast-sampling arguments to avoid NaN errors or crashing. If --upcast-sampling works as a fix with your card, you should have 2x speed (fp16) compared to running in full precision.
To run the server, run the following:
docker compose up -d
Download compatible stable diffusion models (*.safetensors) to the models/Stable-diffusion directory.