I run qwen3.8 on 5060ti 16G RAM. If you can do with Q3 and 64k of context. It works great at 25t/s