For your setup, do you have both 3090's in parallel for the inference of the model?