RTX 4090 + llama.cpp + Qwen3.6 27B MTP for Pi coding agent — is this config reasonable?

Article automatically generated from technical news.

Hi everyone, I’m setting up a local coding-agent workflow on a Windows machine and I’d like some feedback on my llama.cpp configuration. Hardware / setup: GPU : RTX 4090, 24 GB VRAM CPU : Intel i9-13900K Backend : llama.cpp / llama-server Agent : Pi coding agent Model : unsloth/Qwen3.6-27B-MTP-GGUF Quant : Qwen3.6-27B-Q5_K_M.gguf OS : Windows, running from PowerShell/SSH My goal is to use Pi as a coding agent on medium-to-large projects. I care more about stability and go