My first impressions on ROCm and Strix Halo
Summary
The post offers practical first impressions of using ROCm with the Strix Halo GPU, including OS setup, BIOS/GRUB memory configurations, and the challenges of getting PyTorch with ROCm working via UV. It also documents running Llama.cpp and Qwen-3.6 with ROCm using containerized workflows (Podman), plus a concrete OpenCode integration example. Overall, it provides hands-on guidance for Linux users adopting ROCm hardware stacks and ML tooling.