I built a native macOS menu bar app to monitor NVIDIA GPUs across remote Linux servers.
One problem I kept running into was how monitoring tools handle failures. If SSH goes down, that doesn't mean eight GPUs have failed. And if nvidia-smi stops responding, we shouldn't report its last known values as current.
GPU Watch keeps SSH connectivity, GPU query results, and data freshness separate.
It's currently available as a source build for macOS.
Built at SIAM.AI, and I'd appreciate feedback from anyone managing remote GPU servers.