AMD Support.

[Nacon]

We are able to run SD on AMD via ONNX on Window system. So I'm wondering how likely can we see WebUI supporting this?

I do realize it won't able to use the upscaler, but would be ok if it didn't come with that. It's the interface, in-painting and others that I'm interested in.

[Waiwei8k]

I got an AMD card running using these instructions: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs . Is that what you were asking for?

[ChaseFlorom]

I also hasn't found a solution...Surely there's people with AMD and Windows who want to run it?

[TheRealBlackNet]

There are people! I try to run it on a second computer with a AMD card but for the moment i use the option with the full precision mode so it runs on the CPU and each pic takes 3 minutes. Sad there are only tutorials for the cuda\commandline version and none for the webui.

[Waiwei8k]

That's for Linux. I was asking for Window version.
Fair point, though you could perhaps install Linux to an external drive and just use that for SD. Not a fix, but might get you going whilst you wait.

[cdrewing]

Well, I am running on Linux since Ubuntu Feisty Fawn (i.e. 2007). But this is wayyy to advanced for me! I'd really appreciate to find a step by step guide somewhere.

[f1am3d]

@Nacon That actually also works for Windows, I checked. But the benefit of ONNX that it gives the same performance on AMD GPU as on nVidia ones.

[Nacon]

Check this out... https://github.com/azuritecoin/OnnxDiffusersUI

[cdrewing]

Will this work easily? I am the guy from above.

[ClashSAN]

Check this out... https://github.com/azuritecoin/OnnxDiffusersUI

Anyone know, Is this optimized for fast inference on cpu?
I know people will compare openvino vs onnxruntime for cpu inference only. Openvino maybe slightly beats the latter but has 3 stages of model conversion: ckpt>diffusers>onnx>IR
The onnx only needs two, am I right? ckpt>diffusers>onnx
I maybe will just test it myself

[linonetwo]

https://github.com/lshqqytiger/stable-diffusion-webui-directml works on Windows + AMD GPU

Tested on Win10, mine is RX 480, and is 4x faster (2.5-3s/it) than CPU (10-12s/it), not much, but at least faster...

[mziada]

Hi @linonetwo, I also have RX 480 and Windows 11 (tried on Windows 10), I didn't manage to get it working I am not sure why.
I folloed the guide from AMD here https://community.amd.com/t5/ai/updated-how-to-running-optimized-automatic1111-stable-diffusion/ba-p/630252
Could you please let me know what did you do or what guide did you follow?
Basically, I still get the Cuda error every time I run it!!

[marcus-n-andersson]

@mziada Have you gotten it working? I also get the CUDA error together with some other issues as well.

[mziada]

Yes, add --direct-ML arguments in the batch file