Closed yesbroc closed 1 year ago
I am getting 3min 2 sec for Euler A, 20steps
1024x1024 (3050)
Drivers are Pre-vram offload update.
--always-batch-cond-uncond
is slowing it down, --opt-split-attention-v1 produces slightly larger images but is slightly slower.
I use --lowvram --opt-sdp-no-mem-attention
anyway to use the refiner without any fancy extensions?
depends if auto wants to support it in the main repo. He said earlier someone could make an extension, and sd-webui-refiner works with one click ~6gb needed for now
BTW enable your pagefile and use the updated sdxl model, to run vae in fp16
Is there an existing issue for this?
What happened?
comfy ui: ~260seconds 1024 1:1 20 steps a1111: 3600 seconds 1024 1:1 20 steps
Steps to reproduce the problem
N/A
What should have happened?
N/A
Version or Commit where the problem happens
version: v1.4.1-201-g14cf434b
What Python version are you running on ?
Python 3.10.x
What platforms do you use to access the UI ?
Windows
What device are you running WebUI on?
Nvidia GPUs (RTX 20 above)
Cross attention optimization
xformers
What browsers do you use to access the UI ?
Microsoft Edge
Command Line Arguments
List of extensions
Additional information
i have an rtx 3050 ti (mobile 4gb), unfortunate ofc. still dk why the speed difference is so huge