LocalLLaMA

Now that MTP is merged… What’s the best outputs you’re getting on Qwen 3.6 35B on 2x3090s?

We've got great outputs for 27B via club 3090, but what about those of us who love the blazing speed of 35B on dual 3090s? I was getting 1500 p/p and 120 t/g with split layers, but MTP slowed it down to 80 t/g when I tested last week. I'm stic…