Huge performance degradation in latest update

[whab]

My system (64 RAM, RTX 3080Ti 12GB VRAM) shows a huge performance degradation with the latest update (on Aug 10 in Pinokio) versus the original installation (on Aug 8): the Schnell model is now 12 x SLOWER to generate the same image, 240 secs instead of the original 20 secs. The inference is now even slower than the original implementation (https://github.com/Stability-AI/stable-fast-3d.git)... Note that this performance degradation is not specific to my Pinokio installation as it also occurs with a clean manual installation.

[cocktailpeanut]

could you try deleting this line and retry? https://github.com/pinokiofactory/flux-webui/blob/main/app.py#L80

i've asked several people with your spec and it seems to be mostly working fine, but might be because of that line where it uses cpu offload (which makes it possible to run on low vram machines)