Stable Diffusion Not Using AMD GPU (Slow, CPU Only)

How to fix it

Try these in order — the most common fix is first.

1. 1. Windows: DirectML fork

   Copy steps
   1. Install the community DirectML fork of the web UI (lshqqytiger/stable-diffusion-webui-directml)
   2. Add --use-directml to COMMANDLINE_ARGS
   3. Update the AMD driver
   Did this fix it? Yes No

2. 2. Linux: ROCm

   Copy steps
   1. Check that your card is on AMD's ROCm support list
   2. Install the ROCm build of PyTorch in a fresh venv
   Did this fix it? Yes No

3. 3. If it stays slow

   Copy steps
   1. Use --medvram on 8 GB cards
   2. Expect the first image to be slower while shaders compile
   3. Consider a cloud GPU or an NVIDIA card
   Did this fix it? Yes No

Symptoms

'Torch is not able to use GPU', GPU usage near 0% while the CPU is busy, or images take minutes.

Why it happens

• CUDA build of PyTorch installed on an AMD card
• Stock A1111 does not support AMD on Windows
• ROCm is Linux-oriented and supports only some cards
• Shader compilation makes first generations slow

This guide also covers

• stable diffusion amd gpu
• stable diffusion amd slow
• stable diffusion directml not using gpu

The same problem in other apps

All slow, laggy or resource-hungry guides →

• Why Is Midjourney So Slow (Midjourney)

Other Stable Diffusion errors

• Stable Diffusion CUDA Out of Memory Error
• Stable Diffusion "Torch is not able to use GPU" Error
• Stable Diffusion Black Images and NaN Error
• Stable Diffusion Model Failed to Load, Exiting
• Stable Diffusion Very Slow on Mac (M1, M2, M3)
• Stable Diffusion xformers Error After Update
• Stable Diffusion "Couldn't install torch" Error
• Stability AI API 402 Insufficient Credits Error