Introduction
My laptop configuration - ryzen 5600h, rtx 3050ti, 8gb ram + 8gb zram.
On cable gpu soncumes 60w, on battery 30w.
I’ve used this WebUI to work with stable diffusion, encountered some problems. Like I’ve needed to install xformers manually via pip, gfpgan the same way and to copy real-esrgan into the default folders. Also I 've needed to use this argument --skip-torch-cuda-test because PyTorch wasn’t able to identify my GPU though it’s supported by it.
And I’ve remade webui.sh to just this one-liner - python3.10 launch.py --xformers --medvram.
Test
4 images, 77 steps, ddim, 512x512, ginger cat in Ilya Repin artstyle.
Generation time - 1m 19s.
- Also, nvtop showed that besides usage of the dGPU there was a high usage of the iGPU, which I thought is not supported.
