What the lab's GPUs are doing right now. Two nodes, eight NVIDIA RTX 6000 Ada cards, sampled once a second and published here about once a minute. Jobs are research runs — generative models for mobility, LLM/VLM fine-tuning, trajectory and traffic-state work.
The lab's orchestrator samples each GPU once a second over NVML. It is not reachable from the public internet, so a publisher on the lab side writes a stripped-down snapshot — utilization, memory, temperature, power — to a public data repo roughly every minute, and this page reads that file. Nothing on this page queries the lab network, and no job names, user names, host names, or data ever leave it.
“Busy” means a card is at 20% utilization or above at the moment of the snapshot. GPU-hours are utilization-weighted: a card at 50% for two hours counts as one GPU-hour. History is accumulated from the moment the publisher started, so the stated window is the real one, not an assumed full day.