Picomon 0.2.0: From AMD Crash Fix to GPU Monitoring That Doesn't Suck (omarkama.li)

Picomon 0.2.0 has been released, significantly enhancing GPU monitoring capabilities for users working with AMD and NVIDIA hardware, as well as Apple Silicon. Initially created to address challenges with AMD's `amd-smi` tool—which lacked visibility into training efficiency and caused random crashes—the new version is a complete overhaul that introduces multi-vendor support, delivering a user-friendly, live-updating text user interface (TUI) built using the Textual library. Users can access real-time metrics like memory usage, power draw, and utilization in an organized manner, making it suitable for both headless servers and standard setups. This update is particularly relevant for the AI/ML community, as traditional GPU monitoring tools are often limited by vendor locks and can be overly complex for basic tasks. The latest iteration of Picomon eliminates these barriers by allowing seamless integration across different platforms with minimal dependencies. The addition of a "Rig Card" feature encourages knowledge sharing among machine learning engineers, allowing them to display and compare system configurations easily. The tool’s lightweight nature ensures that it runs efficiently in tmux sessions, making it an appealing option for anyone focused on effective resource management during model training. Users can install it via pip, enhancing accessibility and practicality for AI practitioners.