An execution provider (EP) selects which kernels a runtime dispatches for model operations on target hardware. In ONNX Runtime, providers connect the serving runtime to backends such as CUDA, TensorRT, and OpenVINO. section 9.1
A provider without kernels for the target data type can cause a silent FP32 fallback. Verify the executing data type on the deployment host; a quantized model artifact alone does not prove quantized execution. section 9.1 section 9.6
Figure 9.1 Hardware selects the runtime family, and the execution provider selects which kernels dispatch. The leaf node is the production artifact. The two routes to TensorRT and OpenVINO, standalone versus via ORT, trade peak latency against a unified serving surface across heterogeneous fleets. Branches marked Chapter 10 and Chapter 11 point to regimes covered there.

Trace how hardware determines the runtime family, the provider determines kernel dispatch, and the deployment route trades peak latency against a unified serving interface. section 9.1

