Installing Stable Diffusion web UI on AMD Radeon RX 5500M

This tutorial has been tested on AMD RX 5500M gpu and is working properly.

Prerequisites

• Linux distro: Arch
• Hardware
  • AMD Radeon RX 5500M or any Navi 14 based GPU
• Lots of patience!
• AUR package manager: yay, paru etc.

Installing Software

1. ROCm libraries and dependencies

NOTE: This installation is gonna take a lot of time

  yay -S rocm-hip-sdk rocm-opencl-sdk rocminfo rocm-smi-lib

Replace yay with your favorite AUR package manager

• ROCm HIP: A Langauge similar to CUDA by Nvidia.
• ROCm OpenCL: Useful for enchacing run time on cpu based models like (GGFU/GGML)
• rocminfo: To verify if ROCm is working.
• rocm-smi: An efficient way of looking at GPU statistics

Failed to install? run the following command below

  HSA_OVERRIDE_GFX_VERSION=10.3.0 yay -S rocm-hip-sdk rocm-opencl-sdk rocminfo rocm-smi-lib

This makes the installation compatible to gfx1030 which for some reason also works on our device¹

1.1 Add few environment variables

Add these two in your .bashrc or .zshrc

export HSA_OVERRIDE_GFX_VERSION=10.3.0
PATH=$PATH:/opt/rocm/bin

2. Verify ROCm installation

  /opt/rocm/bin/rocminfo

This should output some long information like the below output

  ROCk module is loaded
  =====================
  HSA System Attributes
  =====================
  Runtime Version:         1.1
  System Timestamp Freq.:  1000.000000MHz
  Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
  Machine Model:           LARGE
  System Endianness:       LITTLE

  ========== HSA Agents
  ==========

  ....


  *******
  Agent 2
  *******
    Name:                    gfx1012
    Uuid:                    GPU-XX
    Marketing Name:          AMD Radeon RX 5500M
    Vendor Name:             AMD
    Feature:                 KERNEL_DISPATCH
    Profile:                 BASE_PROFILE
    Float Round Mode:        NEAR

  ...

If you found something like this with your GPU name correctly displayed, then ROCm is installed!

3. Installing PyTorch

First, check if your cpu supports AVX2².

grep avx2 /proc/cpuinfo

If there is no output then your CPU doesn’t support AVX2, so skip the next step.

Install Optimized AVX2 Pytorch

pacman -S python-pytorch-opt-rocm

Install for normal PyTorch (For non AVX2 CPUs)

pacman -S python-pytorch-rocm

Install torch vision

yay -S python-torchvision-rocm

4. Verify PyTorch installation

python -c 'import torch; print(torch.__version__)'

If this command fails then you would need to reinstall PyTorch again (Follow the previous step).

5. Installing Stable Diffusion UI

Clone the repo

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

Setup venv

Copy-pasta’d from Stable diffusion webui wiki

python -m venv venv --system-site-packages
source venv/bin/activate
pip install -r requirements.txt

Run the UI

source venv/bin/activate
./webui.sh --precision full --no-half --upcast-sampling --opt-sub-quad-attention --lowvram --disable-nan-check

The above command is a result of me fixing multiple errors which I’ve faced on my Radeon RX 5500M, feel free to experiment it yourself.

Place your safetensors in models/Stable-diffusion/

Results!

Model used: ZHMix-Dramatic

Note: All of these images are generated and I have no rights or licenses over these images

Some useful links

Hip out of memory error
GPU Specs
RX Series information
Software support list

References

1. Refer this repo ↩︎

2. stackoverflow ↩︎