Disty0
Open a CMD windows in SDNext folder and run these to start SDNext instead: ``` set HSA_OVERRIDE_GFX_VERSION=11.0.2 webui.bat --use-rocm ```
Unfortunately Intel does have general issues with offloading and moving models between CPU and GPU. But this issue in particular doesn't happen on A770 and B580. Not much we can...
IPEX is deprecated, use PyTorch XPU. PyTorch XPU had some support for multi gpu but i don't have a multi gpu environment so i can't confirm anything. Also if you...
Did some more experimenting. Prior model woks fine with IPEX 2.0 but Decoder model fails. Both models fails with IPEX 2.1. IPEX 2.0: (Prior Output Preview / Final Decoder Output)...
This one still happens with IPEX 2.1.20+xpu.
> Can you show me the image you get without corruption on v2.0.120+xpu? Final image is corrupted on all of them. v2.0.120+xpu does generate the intermediate prior stage latent outputs...
Still happens on IPEX 2.3
Stable Cascade works fine on PyTorch 2.5 XPU from PyTorch test branch.
> Stable Cascade works fine on PyTorch 2.5 XPU from PyTorch test branch. But still getting random NaNs that doesn't happen on CPU or other GPU vendors.
Seems to be fixed in IPEX 2.5.10+xpu wheels too, but still the same random NaN behavior as PyTorch 2.5 XPU wheels.