My SDXL workflow was often spilling over my RTX 3080’s 10GB of VRAM into system RAM. I’ve since tweaked my workflow to prevent that, as the slowdowns are dramat

My SDXL workflow was often spilling over my RTX 3080’s 10GB of VRAM into system RAM. I’ve since tweaked my workflow to prevent that, as the slowdowns are dramatic. Tiled VAE encode/decode was a big help.