I was all set on doing the 4x 3090 route but with the current releases of qwen 3.5 and gemma 4. I am having second doubts. 96gb of vram seems to be in a weird spot where it not enough to run larger models and more than needed for the mid models. What are you running as your main model?
