gpu-hot

Why we included this project

GPU Hot gives you a browser page showing what your NVIDIA cards are doing right now, so you don't need to SSH into each box to read nvidia-smi. It reports utilization, memory, temperature, power draw, clocks, PCIe info, and throttle status at sub-second intervals, and it lists active processes with PIDs so you can tell which training job is hogging a GPU. The same Docker image works for a single machine and for a fleet: run it on each GPU server, then point a hub instance at those nodes and watch the whole cluster from one dashboard, with no extra per-server setup beyond the container. That makes it a practical pick for ML engineers and ops folks who want to catch a failing card before it derails a long training run, without adopting a heavyweight monitoring stack.