Developer Runs Local AI Models on AMD RX 9060 XT via WSL and ROCm on Windows

A developer and gamer successfully set up a local AI environment on a personal Windows desktop using an AMD Radeon RX 9060 XT GPU with 16 GB VRAM. The setup involved running Ubuntu 24.04 under Windows Subsystem for Linux (WSL), then installing Ollama along with its dedicated ROCm backend to enable GPU-accelerated inference. A key step was installing librocdxg version 1.2.2, which resolved a GPU detection issue present in the earlier 1.2.0 release, and setting the HSA_ENABLE_DXG_DETECTION environment variable to expose the GPU through the DXG interface. Ollama was configured as a systemd service within WSL, allowing it to run persistently and optionally be accessed from outside the WSL environment. The walkthrough demonstrates that AMD gaming hardware can be repurposed to run large language models locally without requiring dedicated server infrastructure.