Select a workload to see how an NVIDIA GPU SM and a Google TPU systolic array light up. Metrics are derived from public architecture guides and MLPerf submissions so you can compare utilization, memory pressure, and FLOPs/Watt trade-offs.