ONNX on Different Hardware: Why a Successful Launch Doesn't Prove GPU Acceleration

An application starts, an ONNX model loads and the output looks correct. Does that prove the workload...