The japanese did some more modern benchmarks for the latest gpus. I am surprised intel is so damn good. Seems to be a good option if anyone is in the market for non-nvidia
https://chimolog.co/bto-gpu-stable-diffusion-specs/

>>103918677
You don't need that much steps 4-6 should be fine. From looking at the flow in the ksampler advanced the start and end step for both have to add up to the total steps at the top.
So lets say you have total steps 6 and you want half of it on the high model and half on the low model you would set first ksampler 0 as start step and 3 as end. Then on the second you would keep steps the same as before (6) and have 3 as start step and just something really big as the end like 10000000 (so you don't have to keep changing it if you change the total steps later).