Does Automatic1111 actually use GPU acceleration on Mac M Chip?

[haihaict]

1. I am saw my webui "Launching Web UI with arguments: --no-half --upcast-sampling --skip-torch-cuda-test --upcast-sampling --no-half-vae --use-cpu interrogate"
2. but at the same time, everytime I generating picture, I saw this python activity take 95% of GPU.
3. I saw a .py file in venv, which trying to call METAL API accelerator, but I think it doesn't work. Since I believe 99% M1 mac users of AUTOMATIC1111 are running on a rosetta-translated virtual python environment, which does not directly support GPU accelerating by using METAL.

Q1: Why is my automatic1111 still dominating my GPU, since it launched with only using CPU?
Q2: What is the best launching cmd for a MAC M-chip user?

[arcelx]

I came across your post researching similar setups for Mac OS silicon m2. It requires a bit more information to answer your post, but honestly, the best way to go about this is to test the time / speed of automatic1111 by generating the same process but with different launch arguments each time. Note down the speed / time it takes to perform the operation for each set of launch argument.

I found that the fastest for me was:
“—skip-torch-cuda-test —opt-split-attention”
or
“ —skip-torch-cuda-test —opt-split-attention —medvram”